TinyGltfImporter: backport test changes and minor fixes by pezcode · Pull Request #106 · mosra/magnum-plugins

pezcode · 2021-09-26T16:47:02Z

👋 As announced, here are the changes to TinyGltfImporter tests for the upcoming CgltfImporter plugin. Mostly these are about moving out-of-bounds indices into their own files because cgltf checks indices at parsing time. This makes things a bit messier, but it lets us share test files between the two importers as much as possible. I was debating not checking this for cgltf at all, but it would probably be nice to have some tests for the core library, if at least to catch regressions.

The other part are some additional tests I needed for CgltfImporter but thought would be nice to test here, too (even if it just means a FAIL because it's not handled). This includes:

JSON string decoding (handled by tinygltf)
URI decoding (not handled)
duplicate and unordered attributes (handled by tinygltf)
required but unsupported extensions (not handled)
cyclic hierarchies in scenes (not handled)
missing target node in animations (handled by tinygltf, animation is skipped)
invalid camera type (handled by tinygltf)

There's also a small bug fix and some tiny doc changes in there, check out the respective commits for more info.

This explodes when optimizeQuaternionShortestPath is enabled. Fixed in the next commit.

…timizeQuaternionShortestPath is enabled

JSON strings (including keys) are decoded, URIs are not

Narrator: it is not handled

Needed by importers loading buffers on demand, and not at import time. The idea here is to avoid creating new tests in other importers supporting glTF (the upcoming CgltfImporter, possibly AssimpImporter).

Currently not handled

…standard types These attributes are not supported by the core glTF spec. Separate them out, so other importers can test the supported ones only. The idea here is to avoid creating new tests in other importers supporting glTF (the upcoming CgltfImporter, possibly AssimpImporter).

Unlike TinyGltfImporter, some importers may perform index-checking in openData() and would fail to import the *-invalid.gltf fails altogether. Put each OOB check into their own file so this can be tested for each case. The idea here is to avoid creating new tests in other importers supporting glTF (the upcoming CgltfImporter, possibly AssimpImporter).

Moved out of lightInvalid() to avoid errors in light-invalid.gltf in importers performing array size checks in doOpenData(). The idea here is to avoid creating new tests in other importers supporting glTF (the upcoming CgltfImporter, possibly AssimpImporter).

codecov · 2021-09-26T23:36:44Z

Codecov Report

Merging #106 (916d5a0) into master (0e0f8e9) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #106   +/-   ##
=======================================
  Coverage   94.85%   94.86%           
=======================================
  Files         112      112           
  Lines        8857     8872   +15     
=======================================
+ Hits         8401     8416   +15     
  Misses        456      456

Impacted Files	Coverage Δ
.../MagnumPlugins/TinyGltfImporter/TinyGltfImporter.h	`100.00% <ø> (ø)`
...agnumPlugins/TinyGltfImporter/TinyGltfImporter.cpp	`97.52% <100.00%> (+0.03%)`	⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0e0f8e9...916d5a0. Read the comment docs.

…r views The former is not supported and the latter spits out a confusing error message about buffer -1 being out of bounds. Detect both and print helpful error messages.

Tinygltf doesn't check this for us and it triggers CORRADE_INTERNAL_ASSERT_UNREACHABLE in doTexture(). Fixed in the next commit.

This isn't checked by tinygltf

pezcode · 2021-09-27T23:01:24Z

Just finished the last bit of testing for CgltfImporter and added a few more changes, but this should be it 👀

I found a bug and a half that I thought should be fixed. If you want them in another PR to keep this clean, let me know.

mosra

First-class work again, thank you. I went commit-by-commit (excellent job splitting those up, btw) and everything makes sense. Just some minor questions, but it's mostly ready to get merged :)

thought would be nice to test here, too

Definitely, being aware of the plugin shortcomings is a good thing. Right now I'd put the main effort into clgtf, but I can imagine the new checks could eventually get "backported" to the tinygltf importer as well.

If you want them in another PR to keep this clean

No need, the PR is just a transport medium, what matters in the long run is the commits.

mosra · 2021-09-28T09:35:07Z

src/MagnumPlugins/TinyGltfImporter/Test/TinyGltfImporterTest.cpp

+    CORRADE_COMPARE(importer->object3DName(1), "UTF-8: Лорем ипсум долор сит амет");
+    CORRADE_COMPARE(importer->object3DName(2), "UTF-8 escaped: Лорем ипсум долор сит амет");
+    CORRADE_COMPARE(importer->object3DName(3), "Special: \"/\\\b\f\r\n\t");
+    CORRADE_COMPARE(importer->object3DName(4), "Everything: říční člun \t\t\n حليب اللوز");


Haha :D Funny hidden gems here.

mosra · 2021-09-28T09:41:09Z

src/MagnumPlugins/TinyGltfImporter/Test/TinyGltfImporterTest.cpp

              &TinyGltfImporterTest::meshCustomAttributes,
              &TinyGltfImporterTest::meshCustomAttributesNoFileOpened,
+              &TinyGltfImporterTest::meshDuplicateAttributes,
+              &TinyGltfImporterTest::meshUnorderedAttributes,


Interesting, didn't realize this needs to be handled explicitly for cgltf.

If I remember correctly, I briefly considered checking such cases here as well, but given that json.hpp exposes JSON dicts as STL maps (I think?), the non-duplicity and order was implied by that already so I didn't bother. But with cgltf needing an explicit handling it's good to have a check to ensure consistency between the two importers 👍

cgltf doesn't really do... much, besides parsing. Makes it much easier to reason about its inner workings and doesn't bog everyone down in STL madness for things they don't need. It's refreshing to work with even if it means some extra work 👌

Somehow I thought jsmn would sort JSON dicts, but on a second thought it's a non-essential feature, it can be easily done on application side.

mosra · 2021-09-28T09:50:41Z

src/MagnumPlugins/TinyGltfImporter/Test/TinyGltfImporterTest.cpp

                      Containers::arraySize(SingleFileData));

+    addTests({&TinyGltfImporterTest::requiredExtensions,
+              &TinyGltfImporterTest::requiredExtensionsUnsupported});


Narrator: it is not handled

Weak alibi -- historically the plugin was very basic, didn't even have a support for interleaved buffers (!), so checking the extensionsUsed/extensionsRequired fields was a very distant wish. But now that it has a mostly complete support of the core glTF spec and several extensions, implementing those here as well would be a good thing, yep.

Currently I just have a list of supported extensions to check against in CgltfImporter. There might be some extra considerations for handling fallback paths (e.g. the texture extensions), but that's probably a discussion for another PR.

mosra · 2021-09-28T09:56:11Z

src/MagnumPlugins/TinyGltfImporter/Test/required-extensions-unsupported.gltf

+    "extensionsRequired": [
+        "KHR_materials_pbrSpecularGlossiness",
+        "EXT_lights_image_based"
+    ]


glTF Validator tells me that extensionsRequired without the corresponding extensions also in extensionsUsed is invalid. Should we bother checking that in any way? In my opinion what you have here is fine, the meaning is clear, but I wonder if this could lead to some behavioral differences compared to other importers.

I'd say the (non-invalid) test files should definitely pass the validator 👀 I'll fix that.

As for checking that in the importer... I wouldn't bother. We could error out, but once the file is corrected, the import behaviour is the same. It kind of depends on how spec-conformant you want the importer to be and how much validation you want to perform. There are a few cases already where the importer allows 'invalid' (per the spec) files because it really makes no sense to enforce those.

Agree. Yep, it already allows scene-less files and other things not allowed by spec, so why not this as well.

mosra · 2021-09-28T10:04:45Z

src/MagnumPlugins/TinyGltfImporter/Test/mesh-custom-attributes.gltf

+                {
+                    "attributes": {
+                        "_NEGATIVE_PADDING": 4,
+                        "NOT_AN_IDENTITY": 5,


Does the lack of a leading underscore have any meaning or is it just random? Like, cgltf only ever parsing custom attributes that do begin with an underscore? Or, in other words, if the "standard type" section above would have one of them with a leading underscore, would cgltf die on that?

Technically custom attributes without underscore are not allowed, but neither tinygltf nor cgltf care. In CgltfImporter I added a warning "unknown attribute NOT_AN_IDENTITY, importing as custom attribute". But yeah, it could as well go into the "standard type" mesh if it wasn't for the double component type (5130). The warning I mentioned above happens in doOpenData so this works for testing in CgltfImporter, but I'll change it to have one attribute without underscore in "standard types" too. Good catch 😊

unknown attribute NOT_AN_IDENTITY, importing as custom attribute

Okay, you have thought one step ahead again. This is great, thanks!

mosra · 2021-09-28T10:09:13Z

src/MagnumPlugins/TinyGltfImporter/TinyGltfImporter.cpp

-        default: CORRADE_INTERNAL_ASSERT_UNREACHABLE(); /* LCOV_EXCL_LINE */
+        default:
+            Error{} << "Trade::TinyGltfImporter::texture(): invalid minFilter" << s.minFilter;
+            return Containers::NullOpt;


Interesting, for some reason I thought tinygltf did check this, which is why the assert. Or maybe I just looked at the enum values and because there was no "invalid" I assumed it handles this gracefully. Ah well.

Thanks for catching this!

…ore glTF type Just so we have that case in both meshes with and without non-standard types

mosra · 2021-09-28T22:29:06Z

Squashed some of the fixup commits with the originals and merged the series as 0e0f8e9...749b970. Thanks again!

pezcode added 16 commits September 26, 2021 17:58

TinyGltfImporter: docs++

c910243

TinyGltfImporter: fix typo

5b05a77

TinyGltfImporter: test empty animations

1b720f5

This explodes when optimizeQuaternionShortestPath is enabled. Fixed in the next commit.

TinyGltfImporter: fix out-of-bounds read for empty animations when op…

7cf3fef

…timizeQuaternionShortestPath is enabled

TinyGltfImporter: test encoded JSON strings and URIs

1fe6bb3

JSON strings (including keys) are decoded, URIs are not

TinyGltfImporter: test duplicate and unordered attributes

5dac237

TinyGltfImporter: test handling of required extensions

1f806ea

Narrator: it is not handled

TinyGltfImporter: add meshes to buffer-not-found test

fbfda3f

Needed by importers loading buffers on demand, and not at import time. The idea here is to avoid creating new tests in other importers supporting glTF (the upcoming CgltfImporter, possibly AssimpImporter).

TinyGltfImporter: test cycles in scenes

f7f1fe0

Currently not handled

TinyGltfImporter: oops

740af78

TinyGltfImporter: test missing target node in animation

db0d211

TinyGltfImporter: test invalid camera type

6895a54

TinyGltfImporter: fix CI

603bcec

pezcode added 5 commits September 28, 2021 00:46

TinyGltfImporter: handle sparse accessors and accessors without buffe…

911bdd2

…r views The former is not supported and the latter spits out a confusing error message about buffer -1 being out of bounds. Detect both and print helpful error messages.

TinyGltfImporter: test invalid texture sampler values

8c56f5f

Tinygltf doesn't check this for us and it triggers CORRADE_INTERNAL_ASSERT_UNREACHABLE in doTexture(). Fixed in the next commit.

TinyGltfImporter: check for invalid texture sampler values

58a1ef9

This isn't checked by tinygltf

TinyGltfImporter: test buffers without URIs

e60c4f0

TinyGltfImporter: oops

2337f2e

mosra added this to the 2021.0a milestone Sep 28, 2021

mosra reviewed Sep 28, 2021

View reviewed changes

pezcode added 3 commits September 28, 2021 15:28

TinyGltfImporter: test custom attribute without underscore and with c…

c5fe4d9

…ore glTF type Just so we have that case in both meshes with and without non-standard types

TinyGltfImporter: add missing asset object to glTF test files

b1acd6d

TinyGltfImporter: add extensionsUsed to glTF test files

916d5a0

mosra added the changelog mention added label Sep 28, 2021

mosra closed this Sep 28, 2021

pezcode mentioned this pull request Oct 8, 2021

TinyGltfImporter: more backporting #109

Closed

Conversation

pezcode commented Sep 26, 2021

Uh oh!

codecov bot commented Sep 26, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

pezcode commented Sep 27, 2021

Uh oh!

mosra left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mosra commented Sep 28, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

codecov bot commented Sep 26, 2021 •

edited

Loading