Resolve parser defects 2: tests from glslang

[automaticp]

Ran v1.1.2 through a handpicked subset of tests for glslang. Picked out the files that are:

• GLSL (mostly OpenGL, has some minimal Vulkan dialect, but skipped spv.* files for now);
• not ES (removed es keywords in some files that are otherwise valid core);
• do not test for failure or errors, but are expected to compile (commented out some expected-to-fail lines as well). Was only looking for cases where the code is well-formed, but the parser erroneously disagrees.

Don't have time right now to condense to snippets and minimal repros, but I have this output that points to the problematic files:

FAILED(001): comment.frag:11:7: expected ;
FAILED(002): constantUnaryConversion.comp:26:5: expected ,
FAILED(003): constFoldIntMin.frag:6:23: expected ,
FAILED(004): decls.frag:41:17: expected )
FAILED(005): numeral.frag:78:20: expected ,
FAILED(006): preprocessor.edge_cases.vert:3:17: expected a declaration
FAILED(007): preprocessor.function_macro.vert:9:26: expected ;
FAILED(008): stringToDouble.vert:110:36: expected ,
FAILED(009): tokenPaste.vert:18:14: expected ;

FAILED 9 out of 161 files.

Just look for the files with the same names at the link above, the line:column points to the first error occurrence.

• (001): Does not escape a newline in a single-line comment;
• (002): ??? I'd be confused too :woozy_face:, maybe the previous declaration already broke it?
• (003): Non-standard integer suffix as part of the extension;
• (004): Declaration in the while condition. Seems to be legal (see condition and iteration_statement rules);
• (005): Floating-point suffix Lf. Spec only allows f, F, lf, LF, but it's not marked as ERROR in the file. Not sure who's right;
• (006): Multiline comment in a #define :vomiting_face:;
• (007): Newline escape in a macro definition;
• (008): Same as (005);
• (009): Declaration substituted through a macro. If the parser does not keep track of macros, then, well...

Overall pretty clean, mostly newline escapes and (004) are relevant.

[automaticp]

Also should these files be pulled into the samples/ for future regression tests, etc.?

[automaticp]

Actually cannot figure out the license for the test glsl code in glslang. Uh...

[nolanderc]

Oh wow! Nice work!

There will certainly be some issues relates to macros as we don’t expand them, so unless they look like regular function calls/identifiers stuff will go wrong.

Newline escapes is also something I have mostly ignored since it’s rarely used, but I guess macros is an obvious exception.

It would probably be nice to have these somewhere, so feel free to add them! So far I mostly use samples for interactive debugging, but having a more structured approach would be nice.

[nolanderc]

Actually cannot figure out the license for the test glsl code in glslang. Uh...

An alternative is just to extract a minimal example and commit that, it’s what I’ve done so far

[automaticp]

Alright, I'll try to figure something out over the weekend

[nolanderc]

Most of these are fixedin #31.

See #30 for the remaining issue surrounding macros (tokenPaste.vert).