Fix prologue issue

[WhenGryphonsFly]

Supersedes and closes #35.
Supersedes #53.

This PR fixes an issue where some games (e.g., Pokémon Pinball: Ruby and Sapphire) have nonmatching functions under agbcc. This is because these games were compiled with a compiler revision that occasionally pushed LR onto the stack when unnecessary, but otherwise appears to be identical to agbcc.

Unlike #35 and #53, this version of -fprologue-bugfix is defined for OLD_COMPILER/old_agbcc. This is because it is an error to pass an invalid flag to old_agbcc. This causes an issue in pokepinballrs where some files are compiled with agbcc and some are compiled with old_agbcc, but both use the same set of flags. (That is to say, it is easier to define it for OLD_COMPILER than it is to update the Makefile. Especially since I do not know how I would update the Makefile to accomplish this.)

As the Makefile of pokepinballrs has been updated, this is now a copy of #35. Please see below for more information. This has been left open due to the CI issues on #35, but it is functionally the same.

I can be contacted at WhenGryphonsFly#2089 on Discord.

[jiangzhengwenjz]

I don't see why this is better. With #35, when the flag is passed to old_agbcc, it's simply going to be ignored. And I believe pinballrs always uses the same compiler (the one patched by #35, which has agbcc codegen but the same push/pop pattern as old_agbcc)?

[WhenGryphonsFly]

Pokepinballrs uses both compilers (the default compiler is agbcc, but some files such as gbplayer.c use old_agbcc.

Also, this is what happens if I use #35's toplev.c instead of this PR's version:

As can be seen, old_agbcc does not ignore the invalid option, it considers it an error.

[jiangzhengwenjz]

Pokepinballrs uses both compilers (the default compiler is agbcc, but some files such as gbplayer.c use old_agbcc.

This is probably a hack, because sometimes agbcc and old_agbcc have the same codegen (except for push/pop) and the guy needs old_agbcc to not generate the push/pop.

As can be seen, old_agbcc does not ignore the invalid option, it considers it an error.

Oh right. I somehow got it backwards when comparing your PR with #35, sorry.

[WhenGryphonsFly]

This is probably a hack, because sometimes agbcc and old_agbcc have the same codegen (except for push/pop) and the guy needs old_agbcc to not generate the push/pop.

You may be correct. I checked all of the files using old_agbcc, and most compiled the same with agbcc. However, there were still two files that differ in code generation with old_agbcc. I will review them further.

[jiangzhengwenjz]

It's reasonable that some files are using old_agbcc. For example, agb_sram is part of the precompiled static library libagbbackup so it can be compiled with different compiler/flags than the source files written by the devs (the same applies to m4a). Basically they have no reason to use different compilers for their own source code...

[WhenGryphonsFly]

Given that agb_sram.c and m4a.c were the two files remaining, I will then assume that yes, they both used old_agbcc. (I was suspecting as much; one function in agb_sram.c was giving very different codegen.)

Which leaves us back at the start; pokepinballrs uses both compilers, and I have very limited knowledge when it comes to Makefiles. I'll try working on the Makefile (with only two files I think I can work something out), but I'm going to leave this up for now. My understanding is #35 is stalled because the PR author never rebased to rerun the CI, and given that I have a pokepinballrs branch that depends on either #35 or this, I do have a vested interest in getting something merged in quickly.

[jiangzhengwenjz]

You can use target-specific variable values. Basically something like this: https://github.com/pret/pokepinballrs/blob/master/Makefile#L154
And you just need to set the flags for the m4a source.

[WhenGryphonsFly]

Okay, I've updated my pokepinballrs branch. I'll add the #ifndef OLD_COMPILER back in, but as stated above I'll leave the PR open because of the CI issues with #35. Whichever of the two gets merged in doesn't matter to me.

[mid-kid]

@jiangzhengwenjz this is not a hack - there's a specific released version of the compiler, with the filename ThumbPatch_en.zip, that matches this exact behavior. It sits somewhere between old_agbcc and agbcc, and there's backup files in the source code that suggest its existence. See mid-kid/arm-000512@79d7d46

[jiangzhengwenjz]

@mid-kid You misread my reply. I was saying ppl's configuration in makefile was a hack.