gcc 9 and 10

For those of you willing to take risks, experimental builds of gcc 9 and 10 can be found here (under assets):

https://github.com/sodero/adtools/releases

I've only tried this with C code, which seems OK. Raziel tried ScummVM, which is C++, and that seems to be OK as well.

@sTix
Wow! Did you apply patches from adtools which were done for early versions of GCC in full? If so, I may try to build Odyssey with it as well!

Dare I ask if you are willing to deal with some issues later ?:)

I also see you disable gcc17++ in gcc9 due to clib2 limitations, but what about newlib ? (basically, most stuff on os4 use newlib, clib2 for vim was mostly as an exception)

edit: and last question, in gcc10 threading disabled, is it mean whole threading (-athread=native and stuff), or c++ threading?

@kas1e

Quote:

For gcc 9 all patches are applied. The only thing that had to go was c++17 (but that obviously isn't a part of the original patchset) due to limitations in clib2, like you've already noticed. I haven't tried newlib but since those sources are hidden somewhere in the woods it's IMHO better to fix the limitations in clib2.

For gcc 10 some 68k improvements and threading related patches had to go temporarily. But for OS4 and C++ <11 it should be alright. In other words, gcc 10 is still WIP. Gcc 9 has been merged in the main repo but I haven't yet opened a PR for 10 since I want to take a closer look at what I've done (and did not do) first since I will probably need help from Sebastian Bauer and it's a bit rude to ask for help without being atleast somewhat clear when trying to explain the problem :)

Quote:

I can always try. It's not my area of expertise though, and tomorrow the day job will absorb most of my time again.

@sTix
Quote:

There is that bad one: https://github.com/sba1/adtools/issues/76

This one needs it for one of the big games I want to make on os4, and it used very a lot in the game's code in all different ways, so fixing the game's code is pain. And it's better to have more things working in GCC, maybe someday someone will take a task of WebKit port, etc and those things will be in need.

I offer everyone 100$ for just fixing this one, but so far no one wants to deal with :)

@kas1e

Quote:

That's a question I can't answer right now :) I need to do some digging.

@kas1e

Quote:

If noone else wants to touch it then I don't dare to either. I've just spent a day with this code base and I understand almost nothing :)

@kas1e

Github won't let me comment on the issue so I will write my findings here for now (It keeps saying "You can't comment at this time."):

Looking at the code in gthr-amigaos-native.c the problem is that init_threadstore() has not been called yet so when __gthread_self() tries to obtain the semaphore &threadstore->sem it leads to a NULL pointer access.

A quick fix would probably be to add the line:
__gthread_once (&threadstore_once, init_threadstore);

in __gthread_self() just after:
__gthread_once (&libs_once, init_libs);

Same as how it is done in the __gthread_create() function.

@salass00

That awnser could be worth 100,00

@Kas1e if this benefits us all and you can compile for example newer Odyssey's.... ill pay half (50,00)

I'm going to try building a fixed gcc 8 first, and if that works out then I will try to generate an updated patch and create a pull request to the sba1/adtools repo.

@Salas00

It works! At least the simple test case from the bug report for sure works: I have "before" and "after" words. Need a bit more tests, but the first 50$ of the promised donations done already :) Once we can be sure that all fine, will drop the second one. Thanks a bunch as always!

I tried it at 8.3.0

ps. That one needs to be in the adtools repo for sure!

@salass00

Quote:

are you using Brave as browser or some browser with ad blocked? This is the culprit.. i have the same problem with brave browser

@afxgroup

I'm using Google Chrome with AdBlockPlus. It hasn't stopped me from commenting before, but I guess it could be the reason.

@Salas00

Seems to works (at least I can build some crap that uses get_id() all over the ways). Some _very_ initial stuff buggy and crashy:

You can find my fixed gcc 8 at:

https://github.com/salass00/adtools

It works with kas1e's test program.

I've also created a pull request to the original adtools repo as I said I would.

@kas1e

It looks like we have c++17 and threading on gcc 9 now. But there seems to be a deployment problem left. As someone who has only done c++ on other platforms, I would expect my 'hello world' to be linked using just -lstdc++. But in this case that will lead to undefined references to __gthred_*. In the archive created by adtools there are thread model object files named gthr-amigaos-native and gthr-amigaos-single. If I link them as well everything works. But I must be missing something here?

@sTix

Bah, you wrote it above kas1e: '-athread=native'. I should pay more attention :)

Well then, looks like it works. But you need to wait a bit more before it ends up in adtools together with salass00s fix. I need to integrate a minor fix in clib2 first.

Will this fix be available in an updated build/package in gcc9/10 somewhere?

@Raziel

I updated https://github.com/sodero/adtools/releases with the latest gcc 9 with c++17. If the clib2 PR takes to long I'll do a build with salass00s fix.

@sTix

Newlib is fine, not really using clib2 fir anything here


Thanks

@Raziel

Feature complete (including the two fixes from salass00) gcc 9 and 10 can be found here:

https://github.com/sodero/adtools/releases/tag/alpha-2

There's still some work to be done on gcc 10, but that's just 68k stuff if I understand things right.