gcc 9 and 10

@kas1e
Quote:

https://github.com/sba1/adtools/pull/113

Once that's in I'll probably make another PR for a couple of minor clib2 changes as well.

@sTix

Nice.
Thank you very much for all your work

@sTix
Thanks a bunch!

Btw, do you know anything about coreutil's "gprof"? I mean we discuss that it does not work anymore (but works some 10 years ago for sure), but I want to move that bug from the dark side, and at least start from something.

So, there is an archive to check:

https://kas1e.mikendezign.com/aos4/tmp/perfomance_monitor_tests.zip


That is an archive with a good test case, and compiled binaries for clib2, newlib and ones for cygwin_x86. For each of them, i build them with -pg and run when need it (i.e. os4 bins on os4 of course, and x86/Cygwin one on x86/Cygwin one). Then analyze resulted gmon.out with gprofs (for os4 os4 version of gprof, for x86 their version), and results in the "result.txt" files.

As can be seen results for newlib and clib2 are empty, while for x86 their necessary info is placed in.

I tried to compare gmon.out from all 3 tests and can see that gmon.out created for clib2 and newlib are different. For clib2 at the end a bit more info placed. For x86 i can see even more information placed at the beginning of a file.

Maybe at end of all things works, just something simple broken... I may try to analyze gmon.out format, to see what exactly we have in our fields, and what x86 version have, so that can be a starting point about fixing gprof (or whatever is broken for now)

If you have anything intersting to say or so, plz speak :)

@kas1e
Quote:

Sounds like it's related to the gdb problems. But it doesn't seem to help to build using an older dwarf version either. It looks like binutils is broken (unless someone has been very busy updating the dwarf format):

BFD: Dwarf Error: found dwarf version '5', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '0', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '768', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '47133', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '200', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '295', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '5464', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '68', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '178', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '1024', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '46852', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '26990', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '260', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '0', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '1029', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '5', this reader only handles version 2, 3 and 4 information.
BFD: Dwarf Error: found dwarf version '1280', this reader only handles version 2, 3 and 4 information.

@sTix
Wait, what is it from and what is it ?:) Seems like we are about to know the roots of issue?

As i understand those gmons.out we have now built with a different dwarf?

From where you got that output ?

@kas1e

You should get those errors if you do like this:

gcc -pg -gdwarf-2 test.c -o test
gprof test gmon.out > _

It's not just a question of dwarf version though, there's obviously something generating garbage data somewhere.

@sTix
So did i understand right (haven study it for real at the moment) that gmon.out format is the same "dwarf" based section which we have inside of the binary if compile it with -gstabs ?

@kas1e
Quote:

I don't know. But it makes sense that gprof needs sort of the same information as gdb to work.

@sTix
Wait, you are on x5000, right? If so, then performance monitor is not working for you at the moment, it's only now hit beta-testing. So don't try to build/test new ones with -pg, they will not work on you in any case till the new kernel is not out.

@kas1e
Yes, I've got an x5000. Good to know that there's a fix on its way. Let's hope that gprof springs to life once it's out. Or perhaps before, you're a beta tester right :) ?

@sTix
Yeah, the performance monitor works on x5k in beta now, but it didn't make gprof work. Because gprof stop working correctly some 10 years ago and didn't work where it works before: pegaos2, a1, etc.

@sTix

iiuc gcc 10.2 patches are in.
Will there be a bleeding edge build soon-ish?

Thanks

@Raziel
Quote:

Yep, here:
https://github.com/sodero/adtools/releases/tag/10.3.0_1

That includes changes in clib2 as well, just minor C11/C2X things. Those changes haven't yet been integrated (so they will be missing if you build adtools yourself right now). I'll create a PR for those tonight unless something unexpected happens though.

@kas1e
Quote:

That's a shame. Do you know the latest working version?

@sTix

Thank you.
Will do a complete scummvm build when I get home

@sTix

ScummVM full built with gcc 11.2.0 (no gcc errors)

Thank you very much

@Raziel

Excellent, happy to hear that.

@sTix

just noticed that SIGPROF is defined in sys/signal.h for newlib but it isn't for clib2, so gprof might only work with newlib compiled programs.

@trgswe

Thanks for noticing. It's not part of the standard (ISO) but I'll add it anyway. I'm afraid it won't help gprof though :(

@trgswe
With -pg i am able to compile both newlib and clib2 versions with default clib2 includes for sure. Another thing is that both do not work the same...


@sTix
Btw, for clib2 to build with -pg support, i use -lprofile
maybe that "libprofile.a" from clib2 can be also a good starting point to checking wtf happens with gprof ?

@kas1e

Your're probably right. It crashes right away for me though. Does it work for you? (in the sense that you can run the executable and get a gmon.out)