[Tofino] Replace bf-asm BUG macros with top-level BUG macros. Remove GNU extensions flags.

[fruffy]

• Remove GNU extensions for bfasm. Seems to produce compilation problems on some recent systems.
• Instead, use the C++20 feature __VA_OPT__ which only emits is argument when there are more variadic args. This seems to compile on all systems, but I need to add some more checks.
• Replace the BUG macros of bfasm with the top-level BUG macros. Also make sure every BUG call actually has a diagnostic. Tried to produce sensible messages for all the calls, but might have slipped in a few places.

TODO: Recheck error messages and consider emitting a output that is different from Compiler Bug?

[fruffy]

@ChrisDodd I couldn't find a way to rewrite the EXPAND_COMMA macros safely, so I resorted to VA_OPT. Also @jafingerhut had trouble compiling the BUG macros so I replaced them and added diagnostics. Let me know if you other suggestions or see something incorrect.

[fruffy]

@ChrisDodd This code touches a lot of files, but most of the changes are just there to add diagnostics to BUG/BUG_CHECK. Could you give this a review? This PR is required to finally merge bf-asm into the studio here: p4lang/open-p4studio#70

I am using VA_OPT although we are not quite C++20 yet. Thankfully, this works on all the systems we test on. It's the only way I have been able to get rid of the gnu-extensions and variadic arguments, which cause compilation problems for various compiler distributions.

[ChrisDodd]

The compiler BUG macros throw an exception rather than printing a message and then exiting. Do we want a top-level exception handler in main (in bfas.cpp) to catch these exceptions and print the message?

[ChrisDodd]

Should the ## be here? In addition, the ) needs to be macro-ized to make paren handling correct. so should be:

#define CLOSE_PAREN   )
#define EXPAND_COMMA_CLOSE(...) __VA_OPT__(, ) __VA_ARGS__ __VA_OPT__(CLOSE_PAREN)

[ChrisDodd]

Or maybe it should always have the ), so the second __VA_OPT__ is not needed. Should be

#define EXPAND_COMMA_CLOSE(...) __VA_OPT__(, ) __VA_ARGS__  )

-- that seems to be what the original code was doing

[ChrisDodd]

It is probably ok like this -- __VA_OPT__() is a noop no matter how you look at it, and the spurious ## is probably just ignored, though a strict reading of the spec suggests that having it here is undefined.

[kfcripps]

@fruffy @ChrisDodd After this PR, I am getting a bunch of errors like this each time I try to locally run ninja clang-format or ninja clang-format-fix-errors:

...
Error reading /local/kfcripps/repos/p4c/.clang-format: Invalid argument
Formatting /local/kfcripps/repos/p4c/tools/ir-generator/methods.cpp
YAML:31:3: error: unknown key 'WhitespaceSensitiveMacros'
  - __VA_OPT__
  ^
Error reading /local/kfcripps/repos/p4c/.clang-format: Invalid argument
Formatting /local/kfcripps/repos/p4c/tools/ir-generator/type.cpp
YAML:31:3: error: unknown key 'WhitespaceSensitiveMacros'
  - __VA_OPT__
  ^
Error reading /local/kfcripps/repos/p4c/.clang-format: Invalid argument
Formatting /local/kfcripps/repos/p4c/tools/ir-generator/type.h
YAML:31:3: error: unknown key 'WhitespaceSensitiveMacros'
  - __VA_OPT__
  ^
Error reading /local/kfcripps/repos/p4c/.clang-format: Invalid argument
clang-format failed. Run make clang-format-fix-errors to fix the complaints.
ninja: build stopped: subcommand failed.

Do you have any idea how to resolve this?

[fruffy]

Which version of clang-format are you using? Seems to be a problem with the config file.
clang-format 18.1.0 is the version the compiler uses.

[kfcripps]

@fruffy I am using clang-format version 18.1.8

[fruffy]

Let me see whether it can be reproduced with that version..

[fruffy]

I can not reproduce it locally and with CI (#5272). Maybe the CMake script picks up a dated clang-format version?

[kfcripps]

@fruffy After doing a clean build, clang-format commands seem to work again. Thanks for checking.