[GRASS-dev] Re: Upcoming 7.2.0: review which addons to move to core

mlennert · October 5, 2016, 8:20am

[sent this from the wrong address, so it didn't get through to the list]

-------- Message d'origine --------
Envoyé : 5 octobre 2016 00:41:20 GMT+02:00

Le 4 octobre 2016 22:55:35 GMT+02:00, "Anna Petrášová" <kratochanna@gmail.com> a écrit :

On Tue, Oct 4, 2016 at 4:22 PM, Markus Metz
<markus.metz.giswork@gmail.com> wrote:

On Tue, Oct 4, 2016 at 5:42 PM, Sören Gebbert
<soerengebbert@googlemail.com> wrote:

Hi,

>
> You are very welcome to write the missing tests for core modules.
>
> However, i don't understand the argument that because many core

modules

> have
> no tests, therefore new modules don't need them. If developers of

addon

> module are serious about the attempt to make their modules usable

and

> maintainable for others, then they have to implement tests. Its

and

> integral
> part of the development process and GRASS has a beautiful test
> environment
> hat makes writing tests easy. Tests and documentation are part of

coding

> and
> not something special. I don't think this is a hard requirement.
>
> There is a nice statement that is not far from the truth:

Untested code

> is
> broken code.

these gunittests only test if a module output stays the same. This

This is simply wrong, please read the gunittest documentation.

but then why does

The gunittest for the v.stream.order addon is an example how its

done:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py

assume certain order numbers for features 4 and 7? What if these

order

numbers are wrong?

Recently I fixed bugs in r.stream.order, related to stream length
calculations which are in turn used to determine stream orders. The
gunittest did not pick up 1) the bugs, 2) the bug fixes.

You can write gunittests that will test every flag, every option,

their

combination and any output of a module. I have implemented plenty of

tests,

that check for correct error handling. Writing tests is effort, but

you have

to do it anyway. Why not implementing a gunittest for every feature

while

developing the module?

My guess for the r.stream.* modules is at least 40 man hours of
testing to make sure they work correctly. That includes evaluation

of

float usage, handling of NULL data, comparison of results with and
without the -m flag. Testing should be done with both high-res

(LIDAR)

and low-res (e.g. SRTM) DEMs.

Tests can be performed on artificial data that tests all aspects of

the

algorithm. Tests that show the correctness of the algorithm for

specific

small cases should be preferred. However, large data should not be

an

obstacle to write a test.

I agree, for tests during development, not for gunittests.

From the examples I read, gunittests expect a specific output. If the
expected output (obtained with an assumed correct version of the
module) is wrong, the gunittest is bogus. gunittests are ok to make
sure the output does not change, but not ok to make sure the output

is

correct. Two random examples are r.stream.order and r.univar.

I am not sure why are we discussing this, it's pretty obvious that
gunittests can serve to a) test inputs/outputs b) catch changes in
results (whether correct or incorrect) c) test correctness of results.
It just depends how you write them, and yes, for some modules c) is
more difficult to implement than for others.

Well, I agree with Markus that unittests are not a panacea and that we should not fall into the trap of thinking that these tests will guarantee that the results of our modules are correct.

However, I do agree that these tests are useful in detecting if any changes to the code change the output, thus raising a flag that the developer has to at least take into account.

I'll try to write some tests for the OBIA tools when I find the time, although I do agree with Markus that it wouldn't be useful to try to write tests that would cover each and every possible corner case...

In the meantime, g.extension is wonderful tool

Moritz

Anna

Markus M

Best regards
Soeren

my2c

Markus M

>
> Best
> Sören
>
>>
>> One thing we could think about is activating the toolbox idea a

bit

>> more
>> and creating a specific OBIA toolbox in addons.
>>
>>> Identified candidates could be added to core once they fulfill

the

>>> requirements above. Would that happen only in minor releases or

would

>>> that also be possible in point releases?
>>
>>
>> Adding modules to core is not an API change, so I don't see why

they

>> can't
>> be added at any time. But then again, having a series of new

modules

>> can be
>> sufficient to justify a new minor release
>>
>>> Or is that already too much formality and if someone wishes to

see an

>>> addon in core that is simply discussed on ML?
>>
>>
>> Generally, I would think that discussion on ML is the best way

to

>> handle
>> this.
>>
>> Moritz
>>
>>
>> _______________________________________________
>> grass-dev mailing list
>> grass-dev@lists.osgeo.org
>> http://lists.osgeo.org/mailman/listinfo/grass-dev
>
> _______________________________________________
> grass-dev mailing list
> grass-dev@lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/grass-dev

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

SoerenGebbert · October 5, 2016, 12:24pm

Hi,

···

2016-10-05 10:20 GMT+02:00 Moritz Lennert <mlennert@club.worldonline.be>:

[sent this from the wrong address, so it didn’t get through to the list]

-------- Message d’origine --------
Envoyé : 5 octobre 2016 00:41:20 GMT+02:00

Le 4 octobre 2016 22:55:35 GMT+02:00, “Anna Petrášová” <kratochanna@gmail.com> a écrit :

On Tue, Oct 4, 2016 at 4:22 PM, Markus Metz
<markus.metz.giswork@gmail.com> wrote:

On Tue, Oct 4, 2016 at 5:42 PM, Sören Gebbert
<soerengebbert@googlemail.com> wrote:

Hi,

You are very welcome to write the missing tests for core modules.

However, i don’t understand the argument that because many core
modules
have
no tests, therefore new modules don’t need them. If developers of
addon
module are serious about the attempt to make their modules usable
and
maintainable for others, then they have to implement tests. Its
and
integral
part of the development process and GRASS has a beautiful test
environment
hat makes writing tests easy. Tests and documentation are part of
coding
and
not something special. I don’t think this is a hard requirement.

There is a nice statement that is not far from the truth:
Untested code
is
broken code.

these gunittests only test if a module output stays the same. This

This is simply wrong, please read the gunittest documentation.

but then why does

The gunittest for the v.stream.order addon is an example how its
done:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py

assume certain order numbers for features 4 and 7? What if these
order
numbers are wrong?

Recently I fixed bugs in r.stream.order, related to stream length
calculations which are in turn used to determine stream orders. The
gunittest did not pick up 1) the bugs, 2) the bug fixes.

You can write gunittests that will test every flag, every option,
their
combination and any output of a module. I have implemented plenty of
tests,
that check for correct error handling. Writing tests is effort, but
you have
to do it anyway. Why not implementing a gunittest for every feature
while
developing the module?

My guess for the r.stream.* modules is at least 40 man hours of
testing to make sure they work correctly. That includes evaluation
of
float usage, handling of NULL data, comparison of results with and
without the -m flag. Testing should be done with both high-res
(LIDAR)
and low-res (e.g. SRTM) DEMs.

Tests can be performed on artificial data that tests all aspects of
the
algorithm. Tests that show the correctness of the algorithm for
specific
small cases should be preferred. However, large data should not be
an
obstacle to write a test.

I agree, for tests during development, not for gunittests.

From the examples I read, gunittests expect a specific output. If the
expected output (obtained with an assumed correct version of the
module) is wrong, the gunittest is bogus. gunittests are ok to make
sure the output does not change, but not ok to make sure the output
is
correct. Two random examples are r.stream.order and r.univar.

I am not sure why are we discussing this, it’s pretty obvious that
gunittests can serve to a) test inputs/outputs b) catch changes in
results (whether correct or incorrect) c) test correctness of results.
It just depends how you write them, and yes, for some modules c) is
more difficult to implement than for others.

Well, I agree with Markus that unittests are not a panacea and that we should not fall into the trap of thinking that these tests will guarantee that the results of our modules are correct.

Then i live in a parallel universe. Simple question: How do you test your software? How do you assure the correct functionality of your software? Why is it impossible to implement your approach of testing in a dedicated gunittest? How do you assure software quality, if you don’t provide tools so that other developers are able to test your software for correctness? Regression tests are not possible then, because the effect of changes in the core libraries can not be easily detected in modules without tests.

Can you explain to me why the developers of the sophisticated software system VTK [1] implement unit and integration tests for all software components to assure the correct functionality of the framework? They didn’t saw the trap? They are delusional to think that tests assure software quality?

Why is test driven development [2] an integral part of agile software development approaches like scrum or extreme programming? They didn’t saw the trap? They are delusional to think that tests assure software quality?

[1] http://www.vtk.org/overview/
[2] https://en.wikipedia.org/wiki/Test-driven_development

However, I do agree that these tests are useful in detecting if any changes to the code change the output, thus raising a flag that the developer has to at least take into account.

I’ll try to write some tests for the OBIA tools when I find the time, although I do agree with Markus that it wouldn’t be useful to try to write tests that would cover each and every possible corner case…

Why is it “not useful” to write tests for all cases the software is dedicated to solve? It is indeed a lot of effort, but it is useful.

In the meantime, g.extension is wonderful tool

Exactly!

Best regards
Soeren

Moritz

Anna

Markus M

Best regards
Soeren

my2c

Markus M

Best
Sören

One thing we could think about is activating the toolbox idea a
bit
more
and creating a specific OBIA toolbox in addons.

Identified candidates could be added to core once they fulfill
the
requirements above. Would that happen only in minor releases or
would
that also be possible in point releases?

Adding modules to core is not an API change, so I don’t see why
they
can’t
be added at any time. But then again, having a series of new
modules
can be
sufficient to justify a new minor release

Or is that already too much formality and if someone wishes to
see an
addon in core that is simply discussed on ML?

Generally, I would think that discussion on ML is the best way
to
handle
this.

Moritz

grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

mlennert · October 5, 2016, 1:20pm

On 05/10/16 14:24, Sören Gebbert wrote:

Hi,

2016-10-05 10:20 GMT+02:00 Moritz Lennert <mlennert@club.worldonline.be
<mailto:mlennert@club.worldonline.be>>:

    [sent this from the wrong address, so it didn't get through to the list]

    -------- Message d'origine --------
    Envoyé : 5 octobre 2016 00:41:20 GMT+02:00

    Le 4 octobre 2016 22:55:35 GMT+02:00, "Anna Petrášová"
    <kratochanna@gmail.com <mailto:kratochanna@gmail.com>> a écrit :
    >On Tue, Oct 4, 2016 at 4:22 PM, Markus Metz
    ><markus.metz.giswork@gmail.com
    <mailto:markus.metz.giswork@gmail.com>> wrote:
    >> On Tue, Oct 4, 2016 at 5:42 PM, Sören Gebbert
    >> <soerengebbert@googlemail.com
    <mailto:soerengebbert@googlemail.com>> wrote:
    >>> Hi,
    >>>>
    >>>> >
    >>>> > You are very welcome to write the missing tests for core modules.
    >>>> >
    >>>> > However, i don't understand the argument that because many core
    >modules
    >>>> > have
    >>>> > no tests, therefore new modules don't need them. If developers of
    >addon
    >>>> > module are serious about the attempt to make their modules usable
    >and
    >>>> > maintainable for others, then they have to implement tests. Its
    >and
    >>>> > integral
    >>>> > part of the development process and GRASS has a beautiful test
    >>>> > environment
    >>>> > hat makes writing tests easy. Tests and documentation are part of
    >coding
    >>>> > and
    >>>> > not something special. I don't think this is a hard requirement.
    >>>> >
    >>>> > There is a nice statement that is not far from the truth:
    >Untested code
    >>>> > is
    >>>> > broken code.
    >>>>
    >>>> these gunittests only test if a module output stays the same. This
    >>>
    >>> This is simply wrong, please read the gunittest documentation.
    >>
    >> but then why does
    >>>
    >>> The gunittest for the v.stream.order addon is an example how its
    >done:
    >>>
    >https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py
    <https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py>
    >>
    >> assume certain order numbers for features 4 and 7? What if these
    >order
    >> numbers are wrong?
    >>
    >> Recently I fixed bugs in r.stream.order, related to stream length
    >> calculations which are in turn used to determine stream orders. The
    >> gunittest did not pick up 1) the bugs, 2) the bug fixes.
    >>
    >>>
    >>> You can write gunittests that will test every flag, every option,
    >their
    >>> combination and any output of a module. I have implemented plenty of
    >tests,
    >>> that check for correct error handling. Writing tests is effort, but
    >you have
    >>> to do it anyway. Why not implementing a gunittest for every feature
    >while
    >>> developing the module?
    >>>>
    >>>> My guess for the r.stream.* modules is at least 40 man hours of
    >>>> testing to make sure they work correctly. That includes evaluation
    >of
    >>>> float usage, handling of NULL data, comparison of results with and
    >>>> without the -m flag. Testing should be done with both high-res
    >(LIDAR)
    >>>> and low-res (e.g. SRTM) DEMs.
    >>>
    >>> Tests can be performed on artificial data that tests all aspects of
    >the
    >>> algorithm. Tests that show the correctness of the algorithm for
    >specific
    >>> small cases should be preferred. However, large data should not be
    >an
    >>> obstacle to write a test.
    >>
    >> I agree, for tests during development, not for gunittests.
    >>
    >> From the examples I read, gunittests expect a specific output. If the
    >> expected output (obtained with an assumed correct version of the
    >> module) is wrong, the gunittest is bogus. gunittests are ok to make
    >> sure the output does not change, but not ok to make sure the output
    >is
    >> correct. Two random examples are r.stream.order and r.univar.
    >
    >I am not sure why are we discussing this, it's pretty obvious that
    >gunittests can serve to a) test inputs/outputs b) catch changes in
    >results (whether correct or incorrect) c) test correctness of results.
    >It just depends how you write them, and yes, for some modules c) is
    >more difficult to implement than for others.

    Well, I agree with Markus that unittests are not a panacea and that
    we should not fall into the trap of thinking that these tests will
    guarantee that the results of our modules are correct.

Then i live in a parallel universe. Simple question: How do you test
your software? How do you assure the correct functionality of your
software? Why is it impossible to implement your approach of testing in
a dedicated gunittest? How do you assure software quality, if you don't
provide tools so that other developers are able to test your software
for correctness? Regression tests are not possible then, because the
effect of changes in the core libraries can not be easily detected in
modules without tests.

Please note that I was speaking about unit tests, here. I don't know how efficient our testing framework is for integration testing ? Maybe we also need to be clearer about what we understand by tests during such discussions ?

Good discussion, though !

Can you explain to me why the developers of the sophisticated software
system VTK [1] implement unit and integration tests for all software
components to assure the correct functionality of the framework? They
didn't saw the trap? They are delusional to think that tests assure
software quality?

Why is test driven development [2] an integral part of agile software
development approaches like scrum or extreme programming? They didn't
saw the trap? They are delusional to think that tests assure software
quality?

[1] http://www.vtk.org/overview/
[2] https://en.wikipedia.org/wiki/Test-driven_development

    However, I do agree that these tests are useful in detecting if any
    changes to the code change the output, thus raising a flag that the
    developer has to at least take into account.

    I'll try to write some tests for the OBIA tools when I find the
    time, although I do agree with Markus that it wouldn't be useful to
    try to write tests that would cover each and every possible corner
    case...

Why is it "not useful" to write tests for all cases the software is
dedicated to solve? It is indeed a lot of effort, but it is useful.

I would say the question is rather, first, whether it is at all possible, and, second, that maybe by thinking that it is, we are too confident in our tests providing information that they really aren't trying to provide.

But I'm no expert whatsoever, on the topic (I am not a computer scientist, just a scientist programming some tools with my very limited capabilities), so I don't want to stretch this discussion out. I do recommend reading this, though:
http://www.rbcs-us.com/documents/Why-Most-Unit-Testing-is-Waste.pdf

I also like the table close to the top of

http://blog.stevensanderson.com/2009/08/24/writing-great-unit-tests-best-and-worst-practises/

(attached as image)

And let's remember that this all started as the question of what should be required for a module to move from addons to core. The question, therefore, is to find the right balance between necessary effort and our desire to offer functionality to users. This also raises the question of why it would be better for a given module to be in core, rather than in extensions. We could also imagine the opposite direction, i.e. move modules from core to extensions to lighten the work load of maintaining core, while still offering the same functionalities.

IMHO, the largest advantage of having a module in core is that when someone changes internal library APIs, then generally they check all of core and modify what needs to, but this is not necessarily the case for extensions...

Maybe we should ask the users of whether this distinction between modules and core and extensions is really relevant for them, or whether most are perfectly happy to just install extensions.

Moritz

(attachments)

ecodiv · October 5, 2016, 2:29pm

Since you are asking , as an user, my main interest is in good documentation and reproducible examples (which I can than also use to see if outputs make sense to me as an user). In that respect there is no inherent difference between core modules and extensions. What is different is that many (most?) of the core functions are accessible through the menu. I personally don’t find that very important, especially with the modules tab giving fairly easy access to extensions, but I can imagine that for new / other users, especially those more inclined to menu-driven applications, this may make a difference.

···

On 05-10-16 15:20, Moritz Lennert wrote:

On 05/10/16 14:24, Sören Gebbert wrote:

Hi,

2016-10-05 10:20 GMT+02:00 Moritz Lennert <mlennert@club.worldonline.be
mailto:mlennert@club.worldonline.be>:

[sent this from the wrong address, so it didn’t get through to the list]

-------- Message d’origine --------
Envoyé : 5 octobre 2016 00:41:20 GMT+02:00

Le 4 octobre 2016 22:55:35 GMT+02:00, “Anna Petrášová”
<kratochanna@gmail.com mailto:kratochanna@gmail.com> a écrit :

On Tue, Oct 4, 2016 at 4:22 PM, Markus Metz
<markus.metz.giswork@gmail.com
mailto:markus.metz.giswork@gmail.com> wrote:

On Tue, Oct 4, 2016 at 5:42 PM, Sören Gebbert
<soerengebbert@googlemail.com
mailto:soerengebbert@googlemail.com> wrote:

Hi,

You are very welcome to write the missing tests for core modules.

However, i don’t understand the argument that because many core
modules
have
no tests, therefore new modules don’t need them. If developers of
addon
module are serious about the attempt to make their modules usable
and
maintainable for others, then they have to implement tests. Its
and
integral
part of the development process and GRASS has a beautiful test
environment
hat makes writing tests easy. Tests and documentation are part of
coding
and
not something special. I don’t think this is a hard requirement.

There is a nice statement that is not far from the truth:
Untested code
is
broken code.

these gunittests only test if a module output stays the same. This

This is simply wrong, please read the gunittest documentation.

but then why does

The gunittest for the v.stream.order addon is an example how its
done:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py
https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py

assume certain order numbers for features 4 and 7? What if these
order
numbers are wrong?

Recently I fixed bugs in r.stream.order, related to stream length
calculations which are in turn used to determine stream orders. The
gunittest did not pick up 1) the bugs, 2) the bug fixes.

You can write gunittests that will test every flag, every option,
their
combination and any output of a module. I have implemented plenty of
tests,
that check for correct error handling. Writing tests is effort, but
you have
to do it anyway. Why not implementing a gunittest for every feature
while
developing the module?

My guess for the r.stream.* modules is at least 40 man hours of
testing to make sure they work correctly. That includes evaluation
of
float usage, handling of NULL data, comparison of results with and
without the -m flag. Testing should be done with both high-res
(LIDAR)
and low-res (e.g. SRTM) DEMs.

Tests can be performed on artificial data that tests all aspects of
the
algorithm. Tests that show the correctness of the algorithm for
specific
small cases should be preferred. However, large data should not be
an
obstacle to write a test.

I agree, for tests during development, not for gunittests.

From the examples I read, gunittests expect a specific output. If the
expected output (obtained with an assumed correct version of the
module) is wrong, the gunittest is bogus. gunittests are ok to make
sure the output does not change, but not ok to make sure the output
is
correct. Two random examples are r.stream.order and r.univar.

I am not sure why are we discussing this, it’s pretty obvious that
gunittests can serve to a) test inputs/outputs b) catch changes in
results (whether correct or incorrect) c) test correctness of results.
It just depends how you write them, and yes, for some modules c) is
more difficult to implement than for others.

Well, I agree with Markus that unittests are not a panacea and that
we should not fall into the trap of thinking that these tests will
guarantee that the results of our modules are correct.

Then i live in a parallel universe. Simple question: How do you test
your software? How do you assure the correct functionality of your
software? Why is it impossible to implement your approach of testing in
a dedicated gunittest? How do you assure software quality, if you don’t
provide tools so that other developers are able to test your software
for correctness? Regression tests are not possible then, because the
effect of changes in the core libraries can not be easily detected in
modules without tests.

Please note that I was speaking about unit tests, here. I don’t know how efficient our testing framework is for integration testing ? Maybe we also need to be clearer about what we understand by tests during such discussions ?

Good discussion, though !

Can you explain to me why the developers of the sophisticated software
system VTK [1] implement unit and integration tests for all software
components to assure the correct functionality of the framework? They
didn’t saw the trap? They are delusional to think that tests assure
software quality?

Why is test driven development [2] an integral part of agile software
development approaches like scrum or extreme programming? They didn’t
saw the trap? They are delusional to think that tests assure software
quality?

[1] http://www.vtk.org/overview/
[2] https://en.wikipedia.org/wiki/Test-driven_development

However, I do agree that these tests are useful in detecting if any
changes to the code change the output, thus raising a flag that the
developer has to at least take into account.

I’ll try to write some tests for the OBIA tools when I find the
time, although I do agree with Markus that it wouldn’t be useful to
try to write tests that would cover each and every possible corner
case…

Why is it “not useful” to write tests for all cases the software is
dedicated to solve? It is indeed a lot of effort, but it is useful.

I would say the question is rather, first, whether it is at all possible, and, second, that maybe by thinking that it is, we are too confident in our tests providing information that they really aren’t trying to provide.

But I’m no expert whatsoever, on the topic (I am not a computer scientist, just a scientist programming some tools with my very limited capabilities), so I don’t want to stretch this discussion out. I do recommend reading this, though:
http://www.rbcs-us.com/documents/Why-Most-Unit-Testing-is-Waste.pdf

I also like the table close to the top of

http://blog.stevensanderson.com/2009/08/24/writing-great-unit-tests-best-and-worst-practises/

(attached as image)

And let’s remember that this all started as the question of what should be required for a module to move from addons to core. The question, therefore, is to find the right balance between necessary effort and our desire to offer functionality to users. This also raises the question of why it would be better for a given module to be in core, rather than in extensions. We could also imagine the opposite direction, i.e. move modules from core to extensions to lighten the work load of maintaining core, while still offering the same functionalities.

IMHO, the largest advantage of having a module in core is that when someone changes internal library APIs, then generally they check all of core and modify what needs to, but this is not necessarily the case for extensions…

Maybe we should ask the users of whether this distinction between modules and core and extensions is really relevant for them, or whether most are perfectly happy to just install extensions.

Moritz

_______________________________________________
grass-dev mailing list
[grass-dev@lists.osgeo.org](mailto:grass-dev@lists.osgeo.org)
[http://lists.osgeo.org/mailman/listinfo/grass-dev](http://lists.osgeo.org/mailman/listinfo/grass-dev)

mlennert · October 5, 2016, 2:46pm

On 05/10/16 16:29, Paulo van Breugel wrote:

On 05-10-16 15:20, Moritz Lennert wrote:

Maybe we should ask the users of whether this distinction between
modules and core and extensions is really relevant for them, or
whether most are perfectly happy to just install extensions.

Since you are asking , as an user, my main interest is in good
documentation and reproducible examples (which I can than also use to
see if outputs make sense to me as an user). In that respect there is no
inherent difference between core modules and extensions. What is
different is that many (most?) of the core functions are accessible
through the menu. I personally don't find that very important,
especially with the modules tab giving fairly easy access to extensions,
but I can imagine that for new / other users, especially those more
inclined to menu-driven applications, this may make a difference.

Thank you, Paulo, for this feedback.

This means that finding a means to more easily integrate addon modules into menus might be a task to keep in mind.

Maybe the startup script could check the .grass7/addons directory and automatically create an 'addons' menu using [1] ? Maybe the mecanism can be changed in a way to just allow to create toolboxes and that these are automatically integrated into the main menu ?

Just brainstorming, here...

Moritz

[1] https://grass.osgeo.org/grass70/manuals/wxGUI.toolboxes.html

SoerenGebbert · October 5, 2016, 2:50pm

2016-10-05 16:46 GMT+02:00 Moritz Lennert <mlennert@club.worldonline.be>:

On 05/10/16 16:29, Paulo van Breugel wrote:

On 05-10-16 15:20, Moritz Lennert wrote:

Maybe we should ask the users of whether this distinction between
modules and core and extensions is really relevant for them, or
whether most are perfectly happy to just install extensions.

Since you are asking , as an user, my main interest is in good
documentation and reproducible examples (which I can than also use to
see if outputs make sense to me as an user). In that respect there is no
inherent difference between core modules and extensions. What is
different is that many (most?) of the core functions are accessible
through the menu. I personally don't find that very important,
especially with the modules tab giving fairly easy access to extensions,
but I can imagine that for new / other users, especially those more
inclined to menu-driven applications, this may make a difference.

Thank you, Paulo, for this feedback.

This means that finding a means to more easily integrate addon modules
into menus might be a task to keep in mind.

Maybe the startup script could check the .grass7/addons directory and
automatically create an 'addons' menu using [1] ? Maybe the mecanism can be
changed in a way to just allow to create toolboxes and that these are
automatically integrated into the main menu ?

Just brainstorming, here...

How about g.extension will create a menu entry based on the modules
keywords when it installs the module? And it will remove the entry when the
module gets removed?

Best regards
Soeren

Moritz

[1] https://grass.osgeo.org/grass70/manuals/wxGUI.toolboxes.html

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

ecodiv · October 5, 2016, 3:14pm

These sound like interesting options. I don’t have a very good idea what would be the best approach, but I would caution against full freedom for the module to install itself in any location in the menu. I am having in mind how some QGIS addons create menu items, including in the top-level menu, resulting in what for me is a more chaotic bloated environment. In that sense, perhaps there could even be an option at installation to prevent the module from creating a menu item.

···

On 05-10-16 16:50, Sören Gebbert wrote:

2016-10-05 16:46 GMT+02:00 Moritz Lennert <mlennert@club.worldonline.be>:

On 05/10/16 16:29, Paulo van Breugel wrote:

On 05-10-16 15:20, Moritz Lennert wrote:

Maybe we should ask the users of whether this distinction between
modules and core and extensions is really relevant for them, or
whether most are perfectly happy to just install extensions.

Since you are asking , as an user, my main interest is in good
documentation and reproducible examples (which I can than also use to
see if outputs make sense to me as an user). In that respect there is no
inherent difference between core modules and extensions. What is
different is that many (most?) of the core functions are accessible
through the menu. I personally don’t find that very important,
especially with the modules tab giving fairly easy access to extensions,
but I can imagine that for new / other users, especially those more
inclined to menu-driven applications, this may make a difference.

Thank you, Paulo, for this feedback.

This means that finding a means to more easily integrate addon modules into menus might be a task to keep in mind.

Maybe the startup script could check the .grass7/addons directory and automatically create an ‘addons’ menu using [1] ? Maybe the mecanism can be changed in a way to just allow to create toolboxes and that these are automatically integrated into the main menu ?

Just brainstorming, here…

How about g.extension will create a menu entry based on the modules keywords when it installs the module? And it will remove the entry when the module gets removed?

Best regards
Soeren

Moritz

[1] https://grass.osgeo.org/grass70/manuals/wxGUI.toolboxes.html

grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

mlennert · October 5, 2016, 3:18pm

On 05/10/16 17:14, Paulo van Breugel wrote:

On 05-10-16 16:50, Sören Gebbert wrote:

2016-10-05 16:46 GMT+02:00 Moritz Lennert
<mlennert@club.worldonline.be <mailto:mlennert@club.worldonline.be>>:

    On 05/10/16 16:29, Paulo van Breugel wrote:

        On 05-10-16 15 <tel:05-10-16%2015>:20, Moritz Lennert wrote:

            Maybe we should ask the users of whether this distinction
            between
            modules and core and extensions is really relevant for
            them, or
            whether most are perfectly happy to just install extensions.

        Since you are asking , as an user, my main interest is in good
        documentation and reproducible examples (which I can than also
        use to
        see if outputs make sense to me as an user). In that respect
        there is no
        inherent difference between core modules and extensions. What is
        different is that many (most?) of the core functions are
        accessible
        through the menu. I personally don't find that very important,
        especially with the modules tab giving fairly easy access to
        extensions,
        but I can imagine that for new / other users, especially those
        more
        inclined to menu-driven applications, this may make a difference.

    Thank you, Paulo, for this feedback.

    This means that finding a means to more easily integrate addon
    modules into menus might be a task to keep in mind.

    Maybe the startup script could check the .grass7/addons directory
    and automatically create an 'addons' menu using [1] ? Maybe the
    mecanism can be changed in a way to just allow to create toolboxes
    and that these are automatically integrated into the main menu ?

    Just brainstorming, here...

How about g.extension will create a menu entry based on the modules
keywords when it installs the module? And it will remove the entry
when the module gets removed?

These sound like interesting options. I don't have a very good idea what
would be the best approach, but I would caution against full freedom for
the module to install itself in any location in the menu. I am having in
mind how some QGIS addons create menu items, including in the top-level
menu, resulting in what for me is a more chaotic bloated environment. In
that sense, perhaps there could even be an option at installation to
prevent the module from creating a menu item.

The current toolboxes mecanism allows for a specific "user modules" menu which automatically contains anything which is in the .grass7/toolboxes/toolboxes.xml file. So if we just put all addons there, this might avoid the chaos you are talking about.

And as g.extension already does the work of reading the module description and keyword, it shouldn't be too complicated to integrate each new module into a toolbox named by the first keyword (raster, vector, imagery, etc) and to use its description as label.

Moritz

SoerenGebbert · October 5, 2016, 8:42pm

Hi,

···

2016-10-05 15:20 GMT+02:00 Moritz Lennert <mlennert@club.worldonline.be>:

On 05/10/16 14:24, Sören Gebbert wrote:

Hi,

2016-10-05 10:20 GMT+02:00 Moritz Lennert <mlennert@club.worldonline.be
mailto:[mlennert@club.worldonline.be](mailto:mlennert@club.worldonline.be)>:

[sent this from the wrong address, so it didn’t get through to the list]

-------- Message d’origine --------
Envoyé : 5 octobre 2016 00:41:20 GMT+02:00

Le 4 octobre 2016 22:55:35 GMT+02:00, “Anna Petrášová”
<kratochanna@gmail.com mailto:[kratochanna@gmail.com](mailto:kratochanna@gmail.com)> a écrit :

On Tue, Oct 4, 2016 at 4:22 PM, Markus Metz
<markus.metz.giswork@gmail.com
mailto:[markus.metz.giswork@gmail.com](mailto:markus.metz.giswork@gmail.com)> wrote:

On Tue, Oct 4, 2016 at 5:42 PM, Sören Gebbert
<soerengebbert@googlemail.com

mailto:[soerengebbert@googlemail.com](mailto:soerengebbert@googlemail.com)> wrote:

Hi,

You are very welcome to write the missing tests for core modules.

However, i don’t understand the argument that because many core
modules
have
no tests, therefore new modules don’t need them. If developers of
addon
module are serious about the attempt to make their modules usable
and
maintainable for others, then they have to implement tests. Its
and
integral
part of the development process and GRASS has a beautiful test
environment
hat makes writing tests easy. Tests and documentation are part of
coding
and
not something special. I don’t think this is a hard requirement.

There is a nice statement that is not far from the truth:
Untested code
is
broken code.

these gunittests only test if a module output stays the same. This

This is simply wrong, please read the gunittest documentation.

but then why does

The gunittest for the v.stream.order addon is an example how its
done:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py
<https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py>

assume certain order numbers for features 4 and 7? What if these
order
numbers are wrong?

Recently I fixed bugs in r.stream.order, related to stream length
calculations which are in turn used to determine stream orders. The
gunittest did not pick up 1) the bugs, 2) the bug fixes.

You can write gunittests that will test every flag, every option,
their
combination and any output of a module. I have implemented plenty of
tests,
that check for correct error handling. Writing tests is effort, but
you have
to do it anyway. Why not implementing a gunittest for every feature
while
developing the module?

My guess for the r.stream.* modules is at least 40 man hours of
testing to make sure they work correctly. That includes evaluation
of
float usage, handling of NULL data, comparison of results with and
without the -m flag. Testing should be done with both high-res
(LIDAR)
and low-res (e.g. SRTM) DEMs.

Tests can be performed on artificial data that tests all aspects of
the
algorithm. Tests that show the correctness of the algorithm for
specific
small cases should be preferred. However, large data should not be
an
obstacle to write a test.

I agree, for tests during development, not for gunittests.

From the examples I read, gunittests expect a specific output. If the
expected output (obtained with an assumed correct version of the
module) is wrong, the gunittest is bogus. gunittests are ok to make
sure the output does not change, but not ok to make sure the output
is
correct. Two random examples are r.stream.order and r.univar.

I am not sure why are we discussing this, it’s pretty obvious that
gunittests can serve to a) test inputs/outputs b) catch changes in
results (whether correct or incorrect) c) test correctness of results.
It just depends how you write them, and yes, for some modules c) is
more difficult to implement than for others.

Well, I agree with Markus that unittests are not a panacea and that
we should not fall into the trap of thinking that these tests will
guarantee that the results of our modules are correct.

Then i live in a parallel universe. Simple question: How do you test
your software? How do you assure the correct functionality of your
software? Why is it impossible to implement your approach of testing in
a dedicated gunittest? How do you assure software quality, if you don’t
provide tools so that other developers are able to test your software
for correctness? Regression tests are not possible then, because the
effect of changes in the core libraries can not be easily detected in
modules without tests.

Please note that I was speaking about unit tests, here. I don’t know how efficient our testing framework is for integration testing ? Maybe we also need to be clearer about what we understand by tests during such discussions ?

Good discussion, though !

I would like to put the GRASS test framework into perspective, since i think that its capabilities are not well known.

The gunittest framework is not about unit tests. It was designed to test all aspects of the GRASS development. This framework allows you to:

Implement unit tests for the Python libraries, their mthods and classes
Implement and run doctests in the source code of the Python libraries
Run integration tests for all modules, checking correct output for almost all datatypes in GRASS (raster, vector, 3D raster, space-time datasets, categories, color definitions, stdout, …). Module tests are IMHO integration tests, since module make use of different library methods and classes and combine them.
Run C-library tests as unit and integration tests. C-library unit and integration tests can either be implemented in C or via ctypes in Python
Run tests on library level, module level or for all libraries and modules in the whole GRASS source tree using a single command, …
Perform regression tests in dedicated test locations, autimatically triggered by a cronjob or a commit
The framework allows you to run all library unit tests, before module integration tests are performed
It creates temporary mapsest to run without problems in production locations
It logs all tests in detail and generates easy to inspect HTML output at runtime, so you can check the progress of the tests and its gradually available results
It allows on the fly mapset creation and deletion
It supports temporary region environments
It support user defined test data for input generation and output validation

These capabilities allow a wide range of tests to be created, covering most aspects of the GRASS development, with the exception of the GUI.
So please no excuses that the gunittest framework is not capable of implementing a test that is required to assure module correctness.

Best regards
Soeren

Can you explain to me why the developers of the sophisticated software
system VTK [1] implement unit and integration tests for all software
components to assure the correct functionality of the framework? They
didn’t saw the trap? They are delusional to think that tests assure
software quality?

Why is test driven development [2] an integral part of agile software
development approaches like scrum or extreme programming? They didn’t
saw the trap? They are delusional to think that tests assure software
quality?

[1] http://www.vtk.org/overview/
[2] https://en.wikipedia.org/wiki/Test-driven_development

However, I do agree that these tests are useful in detecting if any
changes to the code change the output, thus raising a flag that the
developer has to at least take into account.

I’ll try to write some tests for the OBIA tools when I find the
time, although I do agree with Markus that it wouldn’t be useful to
try to write tests that would cover each and every possible corner
case…

Why is it “not useful” to write tests for all cases the software is
dedicated to solve? It is indeed a lot of effort, but it is useful.

I would say the question is rather, first, whether it is at all possible, and, second, that maybe by thinking that it is, we are too confident in our tests providing information that they really aren’t trying to provide.

But I’m no expert whatsoever, on the topic (I am not a computer scientist, just a scientist programming some tools with my very limited capabilities), so I don’t want to stretch this discussion out. I do recommend reading this, though:
http://www.rbcs-us.com/documents/Why-Most-Unit-Testing-is-Waste.pdf

I also like the table close to the top of

http://blog.stevensanderson.com/2009/08/24/writing-great-unit-tests-best-and-worst-practises/

(attached as image)

And let’s remember that this all started as the question of what should be required for a module to move from addons to core. The question, therefore, is to find the right balance between necessary effort and our desire to offer functionality to users. This also raises the question of why it would be better for a given module to be in core, rather than in extensions. We could also imagine the opposite direction, i.e. move modules from core to extensions to lighten the work load of maintaining core, while still offering the same functionalities.

IMHO, the largest advantage of having a module in core is that when someone changes internal library APIs, then generally they check all of core and modify what needs to, but this is not necessarily the case for extensions…

Maybe we should ask the users of whether this distinction between modules and core and extensions is really relevant for them, or whether most are perfectly happy to just install extensions.

Moritz

annakrat · October 5, 2016, 9:10pm

On Wed, Oct 5, 2016 at 10:46 AM, Moritz Lennert
<mlennert@club.worldonline.be> wrote:

On 05/10/16 16:29, Paulo van Breugel wrote:

On 05-10-16 15:20, Moritz Lennert wrote:

Maybe we should ask the users of whether this distinction between
modules and core and extensions is really relevant for them, or
whether most are perfectly happy to just install extensions.

Since you are asking , as an user, my main interest is in good
documentation and reproducible examples (which I can than also use to
see if outputs make sense to me as an user). In that respect there is no
inherent difference between core modules and extensions. What is
different is that many (most?) of the core functions are accessible
through the menu. I personally don't find that very important,
especially with the modules tab giving fairly easy access to extensions,
but I can imagine that for new / other users, especially those more
inclined to menu-driven applications, this may make a difference.

Thank you, Paulo, for this feedback.

This means that finding a means to more easily integrate addon modules into
menus might be a task to keep in mind.

Maybe the startup script could check the .grass7/addons directory and
automatically create an 'addons' menu using [1] ? Maybe the mecanism can be
changed in a way to just allow to create toolboxes and that these are
automatically integrated into the main menu ?

Just brainstorming, here...

Moritz

[1] https://grass.osgeo.org/grass70/manuals/wxGUI.toolboxes.html

Hmm, this is already implemented (several years I would say). Look
into the tab Modules in layer manager. At least we now know, that
nobody here is using it...
Before adding additional menu Addons, I suggest reading
https://trac.osgeo.org/grass/ticket/1742

Anna

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

ecodiv · October 5, 2016, 11:16pm

On October 5, 2016 23:10:27 Anna Petrášová <kratochanna@gmail.com> wrote:

On Wed, Oct 5, 2016 at 10:46 AM, Moritz Lennert
<mlennert@club.worldonline.be> wrote:

On 05/10/16 16:29, Paulo van Breugel wrote:

On 05-10-16 15:20, Moritz Lennert wrote:

Maybe we should ask the users of whether this distinction between
modules and core and extensions is really relevant for them, or
whether most are perfectly happy to just install extensions.

Since you are asking , as an user, my main interest is in good
documentation and reproducible examples (which I can than also use to
see if outputs make sense to me as an user). In that respect there is no
inherent difference between core modules and extensions. What is
different is that many (most?) of the core functions are accessible
through the menu. I personally don't find that very important,
especially with the modules tab giving fairly easy access to extensions,
but I can imagine that for new / other users, especially those more
inclined to menu-driven applications, this may make a difference.

Thank you, Paulo, for this feedback.

This means that finding a means to more easily integrate addon modules into
menus might be a task to keep in mind.

Maybe the startup script could check the .grass7/addons directory and
automatically create an 'addons' menu using [1] ? Maybe the mecanism can be
changed in a way to just allow to create toolboxes and that these are
automatically integrated into the main menu ?

Just brainstorming, here...

Moritz

[1] https://grass.osgeo.org/grass70/manuals/wxGUI.toolboxes.html

Hmm, this is already implemented (several years I would say). Look
into the tab Modules in layer manager. At least we now know, that
nobody here is using it...

I do use it. As I mentioned above, I find it a convenient way to access / find extensions. My point was that others might prefer to have access to these functions through the menu bar. And initially I was one of them, for some reason it took me time to get used using this modules tab, and to realize / internalize that the modules tab is in fact more powerful having the search function.

So I guess it is a matter of preference, which in turn will often depend on what a person is used to. I have seen a similar discussion concerning moving functions to the processing toolbox, but keeping a link to these functions in the menu.

Before adding additional menu Addons, I suggest reading
https://trac.osgeo.org/grass/ticket/1742

But yes, there are other issues to consider, like space on the menu bar, true.

Anna

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

NikosAlexandris · October 5, 2016, 11:39pm

Apologies for removing all but Soeren's last post.

* Sören Gebbert:

I would like to put the GRASS test framework into perspective, since i
think that its capabilities are not well known.
The gunittest framework is not about unit tests. It was designed to test
all aspects of the GRASS development. This framework allows you to:

* Implement unit tests for the Python libraries, their mthods and classes
* Implement and run doctests in the source code of the Python libraries
* Run integration tests for all modules, checking correct output for almost
all datatypes in GRASS (raster, vector, 3D raster, space-time datasets,
categories, color definitions, stdout, ...). Module tests are IMHO
integration tests, since module make use of different library methods and
classes and combine them.
* Run C-library tests as unit and integration tests. C-library unit and
integration tests can either be implemented in C or via ctypes in Python
* Run tests on library level, module level or for all libraries and modules
in the whole GRASS source tree using a single command, ...
* Perform regression tests in dedicated test locations, autimatically
triggered by a cronjob or a commit
* The framework allows you to run all library unit tests, before module
integration tests are performed
* It creates temporary mapsest to run without problems in production
locations
* It logs all tests in detail and generates easy to inspect HTML output at
runtime, so you can check the progress of the tests and its gradually
available results
* It allows on the fly mapset creation and deletion
* It supports temporary region environments
* It support user defined test data for input generation and output
validation

These capabilities allow a wide range of tests to be created, covering most
aspects of the GRASS development, with the exception of the GUI.
So please no excuses that the gunittest framework is not capable of
implementing a test that is required to assure module correctness.

Soeren, could you name some good examples (not to the
manual/s), some real modules with integrated tests? I am much in need
for good code.

By the way, working code does not necessarily mean good code. And vice
versa, good code does not mean that it works the way it should.
Yet, broken but "good" code (cleanly documented, tested as much as possible)
can be easily fixed and probably easier optimised. Working, though
poor/uncommented/untested code, is frequently, if not always, a
big-time-consuming mission.

The question, the way I understand things, is not if and at what extent
we need tests. GRASS-GIS needs them.
However, no matter the effort, tests will almost never be complete.-
That's what I was told by computer scientists, and what I can confirm
with my minimal experiences by doing, for at least two years now, in a
dozen of mini-projects.

Rather, the question is how can we bring, most, of the theory of testing in to
reality, put it in real action, beautify GRASS' code, whether core or
addons. My understanding is that "our" common understanding is to focus
in raising the overall quality level, make it easy for some-one to read
the code, to understand the code almost as if reading simple text, to enable
some-one to improve or adapt a piece of code relatively fast.

Step-by-step. Nothing is going to change overnight.
Not to forget, the community is a mixture of members with a high
diversity degree when looking at coding skills. Everyone should be
invited. And everyone is invited. Experienced developers, are asked
and, with all the respect, (you) carry the responsibility to shed light for
the newcomers in coding for GRASS-GIS.

Allow me please to reiterate what I have repeated indirectly
or directly, some times: please, dear reader/coder/scripter, when you write code, no matter what you write, comment your code,
use longer variable names, don't hesitate to repeat explaining, document
here and there. Test. Use assertions in your code, even for simple,
single functions. Write code that "invites" the reader.

Yes, testing costs a lot of time. But, if comparing it with the time
spent afterwards in debugging bad code, therein lies the wish to have spent the "less"
time right in the beginning.

Testing will boldly save big, valuable time.-
It'll beautify GRASS-GIS and grow the community.
It'll work for us, not against us.

Nikos

mlennert · October 6, 2016, 10:09am

On 06/10/16 01:39, Nikos Alexandris wrote:

The question, the way I understand things, is not if and at what extent
we need tests. GRASS-GIS needs them.
However, no matter the effort, tests will almost never be complete.-
That's what I was told by computer scientists, and what I can confirm
with my minimal experiences by doing, for at least two years now, in a
dozen of mini-projects.

This is true, in my experience, but Markus' remarks (AFAIU) go beyond the question of completeness. One big issue we have with some modules is how to define the 'correct' result. Unless we use a tautological reasoning to say that the current state of the module gives the correct results (thus reducing the signification of the tests to regression tests), there are a series of modules for which no third-party validation data exists.

For example: what is a "correct" segmentation by i.segment ? Or what is the "correct" output of some of r.watershed's results ?

A second issue is that some bugs only appear with large amounts of data (see #3084 for example), but I don't think that it is feasible to test all our modules on multi-GB datasets. And sometimes it is not the size of the data as such, but very specific combinations of data.

This second issue obviously should not stop us from writing tests for less sizeable data. The first issue is a bit more difficult to solve.

Yes, testing costs a lot of time. But, if comparing it with the time
spent afterwards in debugging bad code, therein lies the wish to have
spent the "less" time right in the beginning.

I agree...to a point. However, don't forget that tests are code just like any other code and thus increasing the lines of code by writing all these tests also increases the lines of code that need to be maintained. So, a big yes for testing, but let's be careful and efficient about it and not pretend we can solve everything with testing. And let's be especially wary of the danger cited often in the literature about testing: while thinking about tests is important, don't forget to think about the actual code you write

Moritz

SoerenGebbert · October 6, 2016, 10:18am

Hi Nikos,

···

2016-10-06 1:39 GMT+02:00 Nikos Alexandris <nik@nikosalexandris.net>:

Apologies for removing all but Soeren’s last post.

Sören Gebbert:

I would like to put the GRASS test framework into perspective, since i
think that its capabilities are not well known.
The gunittest framework is not about unit tests. It was designed to test
all aspects of the GRASS development. This framework allows you to:

Implement unit tests for the Python libraries, their mthods and classes

Implement and run doctests in the source code of the Python libraries

Run integration tests for all modules, checking correct output for almost
all datatypes in GRASS (raster, vector, 3D raster, space-time datasets,
categories, color definitions, stdout, …). Module tests are IMHO
integration tests, since module make use of different library methods and
classes and combine them.

Run C-library tests as unit and integration tests. C-library unit and
integration tests can either be implemented in C or via ctypes in Python

Run tests on library level, module level or for all libraries and modules
in the whole GRASS source tree using a single command, …

Perform regression tests in dedicated test locations, autimatically
triggered by a cronjob or a commit

The framework allows you to run all library unit tests, before module
integration tests are performed

It creates temporary mapsest to run without problems in production
locations

It logs all tests in detail and generates easy to inspect HTML output at
runtime, so you can check the progress of the tests and its gradually
available results

It allows on the fly mapset creation and deletion

It supports temporary region environments

It support user defined test data for input generation and output
validation

These capabilities allow a wide range of tests to be created, covering most
aspects of the GRASS development, with the exception of the GUI.
So please no excuses that the gunittest framework is not capable of
implementing a test that is required to assure module correctness.

Soeren, could you name some good examples (not to the
manual/s), some real modules with integrated tests? I am much in need
for good code.

Some examples:

C-Code

Unittest for the linear equation solver in gmath library:

https://trac.osgeo.org/grass/browser/grass/trunk/lib/gmath/test/test_solvers.c

Integration test of the groundwater flow module, that make use of many gmath functions. The result of the simulation is tested against reference values that are based on literature, see the header of the file:

https://trac.osgeo.org/grass/browser/grass/trunk/raster/r.gwflow/testsuite/validation_7x7_grid.py

Python-Code

The PyGRASS unittests are a great example and inspiration for testing:

https://trac.osgeo.org/grass/browser/grass/trunk/lib/python/pygrass/raster/testsuite
https://trac.osgeo.org/grass/browser/grass/trunk/lib/python/pygrass/vector/testsuite

Integration tests:

The v.stream.order uses the PyGRASS vector library and implements two integration tests:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/vector/v.stream.order/testsuite/test_stream_order.py

The temporal module t.rast.univar uses integration tests to validate the combination of the temporal framework, PyGRASS and the functionality of r.univar:

https://trac.osgeo.org/grass/browser/grass/trunk/temporal/t.rast.univar/testsuite/test_univar.py

The module t.rast.aggreagte combines the temporal framework, PyGRASS and r.series. Its functionality is tested in several integration tests:

https://trac.osgeo.org/grass/browser/grass/trunk/temporal/t.rast.aggregate#testsuite

Best regards
Soeren

By the way, working code does not necessarily mean good code. And vice
versa, good code does not mean that it works the way it should.
Yet, broken but “good” code (cleanly documented, tested as much as possible)
can be easily fixed and probably easier optimised. Working, though
poor/uncommented/untested code, is frequently, if not always, a
big-time-consuming mission.

The question, the way I understand things, is not if and at what extent
we need tests. GRASS-GIS needs them.
However, no matter the effort, tests will almost never be complete.-
That’s what I was told by computer scientists, and what I can confirm
with my minimal experiences by doing, for at least two years now, in a
dozen of mini-projects.

Rather, the question is how can we bring, most, of the theory of testing in to
reality, put it in real action, beautify GRASS’ code, whether core or
addons. My understanding is that “our” common understanding is to focus
in raising the overall quality level, make it easy for some-one to read
the code, to understand the code almost as if reading simple text, to enable
some-one to improve or adapt a piece of code relatively fast.

Step-by-step. Nothing is going to change overnight.
Not to forget, the community is a mixture of members with a high
diversity degree when looking at coding skills. Everyone should be
invited. And everyone is invited. Experienced developers, are asked
and, with all the respect, (you) carry the responsibility to shed light for
the newcomers in coding for GRASS-GIS.

Allow me please to reiterate what I have repeated indirectly
or directly, some times: please, dear reader/coder/scripter, when you write code, no matter what you write, comment your code,
use longer variable names, don’t hesitate to repeat explaining, document
here and there. Test. Use assertions in your code, even for simple,
single functions. Write code that “invites” the reader.

Yes, testing costs a lot of time. But, if comparing it with the time
spent afterwards in debugging bad code, therein lies the wish to have spent the “less”
time right in the beginning.

Testing will boldly save big, valuable time.-
It’ll beautify GRASS-GIS and grow the community.
It’ll work for us, not against us.

Nikos

Markus_Metz · October 6, 2016, 8:08pm

On Thu, Oct 6, 2016 at 12:09 PM, Moritz Lennert
<mlennert@club.worldonline.be> wrote:

On 06/10/16 01:39, Nikos Alexandris wrote:

The question, the way I understand things, is not if and at what extent
we need tests. GRASS-GIS needs them.
However, no matter the effort, tests will almost never be complete.-
That's what I was told by computer scientists, and what I can confirm
with my minimal experiences by doing, for at least two years now, in a
dozen of mini-projects.

This is true, in my experience, but Markus' remarks (AFAIU) go beyond the
question of completeness. One big issue we have with some modules is how to
define the 'correct' result. Unless we use a tautological reasoning to say
that the current state of the module gives the correct results (thus
reducing the signification of the tests to regression tests), there are a
series of modules for which no third-party validation data exists.

For example: what is a "correct" segmentation by i.segment ? Or what is the
"correct" output of some of r.watershed's results ?

Very good examples because there is no "correct" output of i.segment
or r.watershed because there is no agreement about what is correct.
Different implementations of image segmentation or 2D hydrological
modelling make different assumptions. Most of these assumptions are
supported by peer-reviewed literature. That means there is no
generally accepted "correct" result, and all we can do is to ensure
that the output of GRASS modules (obviously we assume that the output
matches the expectations of the author of the module) should not
change.

Markus M

A second issue is that some bugs only appear with large amounts of data (see
#3084 for example), but I don't think that it is feasible to test all our
modules on multi-GB datasets. And sometimes it is not the size of the data
as such, but very specific combinations of data.

This second issue obviously should not stop us from writing tests for less
sizeable data. The first issue is a bit more difficult to solve.

Yes, testing costs a lot of time. But, if comparing it with the time
spent afterwards in debugging bad code, therein lies the wish to have
spent the "less" time right in the beginning.

I agree...to a point. However, don't forget that tests are code just like
any other code and thus increasing the lines of code by writing all these
tests also increases the lines of code that need to be maintained. So, a big
yes for testing, but let's be careful and efficient about it and not pretend
we can solve everything with testing. And let's be especially wary of the
danger cited often in the literature about testing: while thinking about
tests is important, don't forget to think about the actual code you write

Moritz

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

annakrat · October 6, 2016, 9:00pm

On Thu, Oct 6, 2016 at 4:08 PM, Markus Metz
<markus.metz.giswork@gmail.com> wrote:

On Thu, Oct 6, 2016 at 12:09 PM, Moritz Lennert
<mlennert@club.worldonline.be> wrote:

On 06/10/16 01:39, Nikos Alexandris wrote:

The question, the way I understand things, is not if and at what extent
we need tests. GRASS-GIS needs them.
However, no matter the effort, tests will almost never be complete.-
That's what I was told by computer scientists, and what I can confirm
with my minimal experiences by doing, for at least two years now, in a
dozen of mini-projects.

This is true, in my experience, but Markus' remarks (AFAIU) go beyond the
question of completeness. One big issue we have with some modules is how to
define the 'correct' result. Unless we use a tautological reasoning to say
that the current state of the module gives the correct results (thus
reducing the signification of the tests to regression tests), there are a
series of modules for which no third-party validation data exists.

For example: what is a "correct" segmentation by i.segment ? Or what is the
"correct" output of some of r.watershed's results ?

Very good examples because there is no "correct" output of i.segment
or r.watershed because there is no agreement about what is correct.
Different implementations of image segmentation or 2D hydrological
modelling make different assumptions. Most of these assumptions are
supported by peer-reviewed literature. That means there is no
generally accepted "correct" result, and all we can do is to ensure
that the output of GRASS modules (obviously we assume that the output
matches the expectations of the author of the module) should not
change.

There is correct result in the sense "as author intended". So you can
use pen and paper and solve very small example. That doesn't mean it
will cover all options, but better than nothing. This is obviously
hard to do if you are not the original author of the algorithm, and
that's the reason we should insist on tests for new modules.

Anna

Markus M

A second issue is that some bugs only appear with large amounts of data (see
#3084 for example), but I don't think that it is feasible to test all our
modules on multi-GB datasets. And sometimes it is not the size of the data
as such, but very specific combinations of data.

This second issue obviously should not stop us from writing tests for less
sizeable data. The first issue is a bit more difficult to solve.

Yes, testing costs a lot of time. But, if comparing it with the time
spent afterwards in debugging bad code, therein lies the wish to have
spent the "less" time right in the beginning.

I agree...to a point. However, don't forget that tests are code just like
any other code and thus increasing the lines of code by writing all these
tests also increases the lines of code that need to be maintained. So, a big
yes for testing, but let's be careful and efficient about it and not pretend
we can solve everything with testing. And let's be especially wary of the
danger cited often in the literature about testing: while thinking about
tests is important, don't forget to think about the actual code you write

Moritz

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev

wenzeslaus · October 6, 2016, 9:34pm

On Tue, Oct 4, 2016 at 4:55 PM, Anna Petrášová <kratochanna@gmail.com>
wrote:

[...]

c) test correctness of results.

It just depends how you write them, and yes, for some modules c) is
more difficult to implement than for others.

On Thu, Oct 6, 2016 at 5:00 PM, Anna Petrášová <kratochanna@gmail.com>
wrote:

There is correct result in the sense "as author intended". So you can
use pen and paper and solve very small example. That doesn't mean it
will cover all options, but better than nothing. [...]

For r.forestfrag, I wrote a test which was based on an example in the
original paper which was computing value of one cell in 3x3 window. It is a
really trivial example which tests 1 number and two other numbers which are
intermediate results. However, by writing it, I actually discovered that
although the results for a large area look good, this small example, which
I was able compute by hand, is failing. After investigation, it turned out
that the error is actually in r.mapcalc. Computing the result outside of
GRASS was crucial in this case. It was an index and I was able to compute
it by hand (and check it with the example in the paper). For some other
modules it could be more difficult and I don't want to compute it without
GRASS for more than one cell even in this case, but it was certainly
possible to write a test of correctness for this module. Note that the test
doesn't show if the (forest fragmentation) index makes sense or if it is
useful, but it shows that the module gives the result it promised.

This is the original test:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/raster/r.forestfrag/testsuite/r_forestfrag_trivial.py

This is the r.mapcalc bug:

https://trac.osgeo.org/grass/ticket/3067

And this is test which specifically shows the bug (would show if it would
be reintroduced):

https://trac.osgeo.org/grass/browser/grass/trunk/raster/r.mapcalc/testsuite/test_row_above_below_bug.py

wenzeslaus · October 6, 2016, 10:47pm

Hi Nikos,

On Wed, Oct 5, 2016 at 7:39 PM, Nikos Alexandris <nik@nikosalexandris.net>
wrote:

could you name some good examples (not to the
manual/s), some real modules with integrated tests? I am much in need
for good code.

I'm not sure if they are good, but here are some examples based on Anna's
test categories.

On Tue, Oct 4, 2016 at 4:55 PM, Anna Petrášová <kratochanna@gmail.com>
wrote:

[...]
gunittests can serve to a) test inputs/outputs b) catch changes in
results (whether correct or incorrect) c) test correctness of results.
It just depends how you write them, and yes, for some modules c) is
more difficult to implement than for others.

a) inputs/outputs

This one tests if the outputs are there (module runs, accepts flags, gives
outputs) and if the output values are in expected ranges ("basic
correctness") and the output are expected data types. It also tests some
stats like sum and variance of the result, but that's more for the category
b).

https://trac.osgeo.org/grass/browser/grass-addons/grass7/raster/r.forestfrag/testsuite/r_forestfrag_ncspm.py

b) result changes

This uses a miniature dataset and results obtained by the module itself. It
tests all values of the result and expects them to match the reference.
This will catch a regression in a module code (or in dependencies).

https://trac.osgeo.org/grass/browser/grass-addons/grass7/raster/r.forestfrag/testsuite/r_forestfrag_xy.py

It assumes that the model worked at the point these were obtained. If that
was not the case, the data must be changed like in this commit:

https://trac.osgeo.org/grass/changeset/68722

c) correctness

Here is a test on a miniature example which was possible to compute by hand
and was part of the original publication.

https://trac.osgeo.org/grass/browser/grass-addons/grass7/raster/r.forestfrag/testsuite/r_forestfrag_trivial.py

Here is a test which uses reference data computed using another software:

https://trac.osgeo.org/grass/browser/grass/trunk/raster/r.slope.aspect/testsuite/test_r_slope_aspect.py?rev=68557#L44
https://trac.osgeo.org/grass/browser/grass/trunk/raster/r.slope.aspect/testsuite/data

Using gdaldem to test r.slope.aspect it is actually not ideal, because
gdaldem uses old r.slope.aspect code. So in this case, it is more a test
against very old version of r.slope.aspect.

This is definitively a creative part because you need to get the results
without using the code you are developing to get the result. For example,
if you have prototype implementation in Python or r.mapcalc and then your
write C code to do the same more effectively with more options, you can use
the prototype to cover part of your outputs. This of course does not give a
guarantee that you did not make the same mistake twice, but it is almost
guaranteed that you will make two separate sets of mistakes and this will
expose them.

It of course never hurts to test also the "basic correctness", e.g. that
the values are all positive or that they are from 0 to 255 and are integers.

d) small parts and dependencies

There is one more category I would add to this list. These are the tests of
building blocks (units) you are using in your module. It can be functions
your wrote but also functions or modules you are using. When writing a C
module you can write a separate module which uses the functions from your
module one by one to generate some testable outputs. These output can be
then tested using the gunittest procedure.

https://trac.osgeo.org/grass/browser/grass/trunk/raster3d/r3.flow/test_main.c
https://trac.osgeo.org/grass/browser/grass/trunk/raster3d/r3.flow/testsuite/r3flow_test.py

Further, you can write a ctypes-based (fully) Python test for the library
functions your are using:

https://trac.osgeo.org/grass/browser/grass/trunk/lib/gis/testsuite/gis_lib_tokenize.py

For Python modules, you can do the same it is just much easier and Soeren
already provided examples for it.

Hope this helps,
Vaclav

mlennert · October 7, 2016, 6:32am

On 06/10/16 23:34, Vaclav Petras wrote:

On Tue, Oct 4, 2016 at 4:55 PM, Anna Petrášová <kratochanna@gmail.com
<mailto:kratochanna@gmail.com>> wrote:

    [...]

    c) test correctness of results.
    It just depends how you write them, and yes, for some modules c) is
    more difficult to implement than for others.

On Thu, Oct 6, 2016 at 5:00 PM, Anna Petrášová <kratochanna@gmail.com
<mailto:kratochanna@gmail.com>> wrote:

    There is correct result in the sense "as author intended". So you can
    use pen and paper and solve very small example. That doesn't mean it
    will cover all options, but better than nothing. [...]

For r.forestfrag, I wrote a test which was based on an example in the
original paper which was computing value of one cell in 3x3 window. It
is a really trivial example which tests 1 number and two other numbers
which are intermediate results. However, by writing it, I actually
discovered that although the results for a large area look good, this
small example, which I was able compute by hand, is failing. After
investigation, it turned out that the error is actually in r.mapcalc.
Computing the result outside of GRASS was crucial in this case. It was
an index and I was able to compute it by hand (and check it with the
example in the paper). For some other modules it could be more difficult
and I don't want to compute it without GRASS for more than one cell even
in this case, but it was certainly possible to write a test of
correctness for this module. Note that the test doesn't show if the
(forest fragmentation) index makes sense or if it is useful, but it
shows that the module gives the result it promised.

This is the original test:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/raster/r.forestfrag/testsuite/r_forestfrag_trivial.py

Nice example, thanks !

This is the r.mapcalc bug:

https://trac.osgeo.org/grass/ticket/3067

And this is test which specifically shows the bug (would show if it
would be reintroduced):

https://trac.osgeo.org/grass/browser/grass/trunk/raster/r.mapcalc/testsuite/test_row_above_below_bug.py

Here we have an example of a test with 70 lines of code which was comitted almost at the same time as the issue was fixed (ok 3 days before), with a one-line fix. Now, it's good to safeguard against this bug reappearing, but I would argue that this example shows that we often only think of relevant tests after a bug appears, and I'm not sure that it is very efficient to write a test at that point, given the low probability of that bug reappearing

But hey, bear with me, I'm learning...

Moritz

Blumentrath_Stefan1 · October 7, 2016, 8:41am

Hi again,

Interesting discussion!

With my user and amateur addon-developer perspective I would conclude that:
- we all agree that tests/unittests are very useful even if not "the silver bullet"
- things one does in "unittests" might also partly be done in the code of the module (e.g. check for a specific state of input or environment variables ...), so a requirement for unittest might have to go hand in hand with a requirement for "maturity" of the module?
- "unittests" can also differ a lot in complexity so it is probably not too easy to say which level of unit tests should be required for a new core module, and the presence of a test alone does not necessarily mean that a module is "well tested"...
- it might be a good starting point to "require" only "simple" unittests for new core modules in that sense that a unit test makes sure that at least examples from the manual work without throwing an error? That would at the same time test code and manual/documentation... There was the idea to use examples from the manual to auto-generate tests, was`nt it?
- with more tests on "higher level" modules, also "lower level functions" (e.g. library) might be covered. This could on the one hand help to trace (side) effect of changes on lower level functions, but on the other hand might require to also update all output-tests which were written based on that bogus library function, would not it?
- like some others here I never wrote a test for my addons, but I would be willing to learn! Here the simple examples Vaclav provided were very helpful, as for me as a rookie the more involved examples and the full documentation of the unittests are hard to understand. Maybe you can add a very minimalistic example to the test documentation that only checks (e.g. based on NC data set) that e.g. "r.example -g input=mymap" works without throwing an error?

And finally, Martins request regarding an RFC for "promoting modules to core" should not drown in the discussion about unittests (even if the latter probably is a precondition to the former)...

Cheers
Stefan

P.S.: For automatized tests of addons there might be a dependency issue, as a test-server would have to be equipped with all possible dependencies in order to be able to run the addons?

-----Original Message-----
From: grass-dev [mailto:grass-dev-bounces@lists.osgeo.org] On Behalf Of Moritz Lennert
Sent: 7. oktober 2016 08:32
To: Vaclav Petras <wenzeslaus@gmail.com>; Anna Petrášová <kratochanna@gmail.com>
Cc: GRASS developers list <grass-dev@lists.osgeo.org>; Markus Metz <markus.metz.giswork@gmail.com>
Subject: Re: [GRASS-dev] Fwd: Re: Upcoming 7.2.0: review which addons to move to core

On 06/10/16 23:34, Vaclav Petras wrote:

On Tue, Oct 4, 2016 at 4:55 PM, Anna Petrášová <kratochanna@gmail.com
<mailto:kratochanna@gmail.com>> wrote:

    [...]

    c) test correctness of results.
    It just depends how you write them, and yes, for some modules c) is
    more difficult to implement than for others.

On Thu, Oct 6, 2016 at 5:00 PM, Anna Petrášová <kratochanna@gmail.com
<mailto:kratochanna@gmail.com>> wrote:

    There is correct result in the sense "as author intended". So you can
    use pen and paper and solve very small example. That doesn't mean it
    will cover all options, but better than nothing. [...]

For r.forestfrag, I wrote a test which was based on an example in the
original paper which was computing value of one cell in 3x3 window. It
is a really trivial example which tests 1 number and two other numbers
which are intermediate results. However, by writing it, I actually
discovered that although the results for a large area look good, this
small example, which I was able compute by hand, is failing. After
investigation, it turned out that the error is actually in r.mapcalc.
Computing the result outside of GRASS was crucial in this case. It was
an index and I was able to compute it by hand (and check it with the
example in the paper). For some other modules it could be more
difficult and I don't want to compute it without GRASS for more than
one cell even in this case, but it was certainly possible to write a
test of correctness for this module. Note that the test doesn't show
if the (forest fragmentation) index makes sense or if it is useful,
but it shows that the module gives the result it promised.

This is the original test:

https://trac.osgeo.org/grass/browser/grass-addons/grass7/raster/r.fore
stfrag/testsuite/r_forestfrag_trivial.py

Nice example, thanks !

This is the r.mapcalc bug:

https://trac.osgeo.org/grass/ticket/3067

And this is test which specifically shows the bug (would show if it
would be reintroduced):

https://trac.osgeo.org/grass/browser/grass/trunk/raster/r.mapcalc/test
suite/test_row_above_below_bug.py

Here we have an example of a test with 70 lines of code which was comitted almost at the same time as the issue was fixed (ok 3 days before), with a one-line fix. Now, it's good to safeguard against this bug reappearing, but I would argue that this example shows that we often only think of relevant tests after a bug appears, and I'm not sure that it is very efficient to write a test at that point, given the low probability of that bug reappearing

But hey, bear with me, I'm learning...

Moritz
_______________________________________________
grass-dev mailing list
grass-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-dev