moldraw2DTest1 failure when building on aarch64

[vfscalfani]

Describe the bug

moldraw2DTest1 fails when building on aarch64 Raspberry Pi OS.

To Reproduce
Hi,

I am trying to build the 2022_09_3 RDKit release on aarch64 (Raspberry Pi OS 64 bit, based on Debian 11).

After building, all tests pass except for one test: moldraw2DTest1. Here are my cmake settings (popcnt off as cmake reported that it was not available on aarch64)

cmake \
 -DRDK_BUILD_INCHI_SUPPORT=ON \
 -DRDK_BUILD_AVALON_SUPPORT=ON \
 -DRDK_BUILD_CAIRO_SUPPORT=ON \
 -DRDK_OPTIMIZE_POPCNT=OFF \
 ..

And here is a snippet of the relevant verbose output of ctest on moldraw2DTest1:

139:  ----------------- Test Github 3305 - change and scale line widths.
139: file testGithub3305_1.png gave hash 255418450U not the expected 316930677U
139: file testGithub3305_2.png gave hash 1774822891U not the expected 3520446560U
139: file testGithub3305_3.png gave hash 970935540U not the expected 3960184199U
139: file testGithub3305_4.png gave hash 667377464U not the expected 411486117U
139: [03:56:23] 
139: 
139: ****
139: Test Assert
139: Expression Failed: 
139: Violation occurred on line 4318 in file /home/pi_test/Downloads/rdkit-Release_2022_09_3/Code/GraphMol/MolDraw2D/test1.cpp
139: Failed Expression: std::regex_search(text, bondMatch, regex) == 1
139: ****
139: 
139: terminate called after throwing an instance of 'Invar::Invariant'
139:   what():  Expression Failed: 
1/1 Test #139: moldraw2DTest1 ...................Child aborted***Exception:  12.19 sec

0% tests passed, 1 tests failed out of 1

Total Test time (real) =  12.22 sec

The following tests FAILED:
	139 - moldraw2DTest1 (Child aborted)

Looking at the test1.cpp file around line 4318, https://github.com/rdkit/rdkit/blob/master/Code/GraphMol/MolDraw2D/test1.cpp, it looked like maybe turning of freetype support may work, however I recieved the same error with disabling freetype in another build:

 -DRDK_BUILD_FREETYPE_SUPPORT=OFF \

Then, since it was failing on a regex_search, I tried to build with boost regex instead, but the same moldraw2DTest1 failed:

 -DRDK_USE_BOOST_REGEX=ON \

This led me to believe that regex is working correctly, and the test is actually failing. I inspected the testGitHub3305_4.svg file (which is what I believe the following regex in test1.cpp is looking at):

test1.cpp, line 4316:

 R"regex(class='bond-6 atom-6 atom-7' d='M 138.\d+,116.\d+ L 141.\d+,116.\d+.* Z' .*FF7F7F.*stroke-width:0.0px)regex");

And here are the 3 lines in the build output testGitHub3305_4.svg that start with class='bond-6:

<path class='bond-6 atom-6 atom-7' d='M 141.9,116.8 L 134.7,129.2 L 133.1,126.5 L 138.7,116.8 Z' style='fill:#FF7F7F;fill-rule:evenodd;fill-opacity:1;stroke:#FF7F7F;stroke-width:0.0px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1;' />
<path class='bond-6 atom-6 atom-7' d='M 140.3,116.8 L 133.9,127.8' style='fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:2.0px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1' />
<path class='bond-6 atom-6 atom-7' d='M 138.1,116.8 L 132.8,125.9' style='fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:2.0px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1' />

I don't think any of those match the expected regex pattern and that this might be a bug.

Many thanks,

Vin

Expected behavior

I expected moldraw2DTest1 to pass.

Screenshots
I have attached the output testGitHub3305_4.svg from the build.

Configuration (please complete the following information):

• RDKit version: 2022_09_3
• OS: Raspberry Pi OS 64 bit
• Python version (if relevant): 3.9.2
• Are you using conda? No
• If you are using conda, which channel did you install the rdkit from? N/A
• If you are not using conda: how did you install the RDKit? building, details above.

Additional context

N/A

[DavidACosgrove]

Hi @vfscalfani,
You do get small numerical differences in the output of the drawing code. The regex is supposed to allow for enough slack that differences below 1 drawing unit don’t cause a failure. In this case it looks like that’s not happening. I’ll take a look at it in the next few days. The crucial thing is that the drawing looks correct. A failing test isn’t a good thing, though.
I don’t have access to a raspberry pi, but I’ll see if the test can be improved.
thanks,
Dave

[vfscalfani]

Hi @DavidACosgrove

Thanks for your reply, that makes sense. I'm happy to test any changes if that would help.

Vin

[DavidACosgrove]

Hi @vfscalfani,

I think I have fixed the test. It was a classic case of "all the right notes, not necessarily in the right order" (obscure UK cultural reference : https://vimeo.com/479336770). The convex hull algorithm that is used to create the rectangle for the highlights was returning the corners of the rectangle in a different order for your file, so the regex failed. It now checks for the correct corners in any order. That might mean a non-rectangular shape will pass the test, but we can cross that bridge later if necessary.
The emended code is at https://github.com/DavidACosgrove/rdkit/tree/Github5899. If you could try it on your raspberry pi that would be great. It seems pointless doing the PR until we know it has fixed your problem.

Best,
Dave

[vfscalfani]

Hi @DavidACosgrove This is great! Yes, I just built with the new code on my Raspberry Pi and moldraw2DTest1 passed (as well as all other tests). That was with Freetype support. I'm running a build now without Freetype support just in case. It does not look like that should matter, but probably good to check anyway. I will report back on that tomorrow (a build and testing takes about 4 hours on the Pi 400). Thanks a lot for fixing this.

Vin

[vfscalfani]

Okay, all tests pass with and without Freetype support. Thanks again @DavidACosgrove

[DavidACosgrove]

Hi Vin, Thanks for running the tests and letting me know. I’ll do the PR. Best, Dave

On Sat, 31 Dec 2022 at 16:02, Vincent F. Scalfani ***@***.***> wrote: Okay, all tests pass with and without Freetype support. Thanks again @DavidACosgrove <https://github.com/DavidACosgrove> — Reply to this email directly, view it on GitHub <#5899 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACGF2FVQJGLWS6PVTXAEBY3WQBKKNANCNFSM6AAAAAATJTAW5U> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- David Cosgrove Freelance computational chemistry and chemoinformatics developer http://cozchemix.co.uk