FPC 3.2.4-rc1 available

[Graeme]

I guess the x86_64-win64 situation (https://gitlab.com/freepascal.org/fpc/source/-/issues/32848) should be mentioned on the release notes?

No, because it is that way since forever.

About 10 years later, and there is still no official 64-bit release for Win64? I know it's simple to compile one yourself - I've done that for over a decade, but come on. 32-bit windows hasn't been a [Microsoft] thing for very long now.

[dbannon]

About 10 years later, and there is still no official 64-bit release for Win64? I know it's simple to compile one yourself - I've done that for over a decade, but come on. 32-bit windows hasn't been a [Microsoft] thing for very long now.

Oh, come on Graham, its over four years since we had an official release for ANY Platform !

Davo

[dsiders]

I guess the x86_64-win64 situation (https://gitlab.com/freepascal.org/fpc/source/-/issues/32848) should be mentioned on the release notes?

No, because it is that way since forever.

About 10 years later, and there is still no official 64-bit release for Win64? I know it's simple to compile one yourself - I've done that for over a decade, but come on. 32-bit windows hasn't been a [Microsoft] thing for very long now.

The Lazarus installer for Windows 64-bit includes a 64-bit compiler.

[marcov]

About 10 years later, and there is still no official 64-bit release for Win64? I know it's simple to compile one yourself - I've done that for over a decade, but come on. 32-bit windows hasn't been a [Microsoft] thing for very long now.

FPC 3.2.4RC1 has a win32/win64 combo installer.  (so win32 + win64 crosscompiler in one installer)

[PeterBB]

I've set up the release candidate on the Arch AUR here;
https://aur.archlinux.org/packages/fpc-rc

Note, to try out with lazarus, it will be necessary to rebuild lazarus first,
otherwise, will get those dreaded "checksum changed" issues!

Cheers,
Peter

[AlexTP]

https://gitlab.com/freepascal.org/fpc/source/-/blob/release_3_2_4-branch/packages/libfontconfig/fpmake.pp?ref_type=heads

File fpmake.pp shows that this 'libfontconfig' package is not compiled for Solaris OS.

Code: Pascal  [Select][+][-]

1.  P.OSes := [linux] + AllBSDOses; // Darwin was tested!

But in the 3.3 Git main, this package IS compiled for Solaris OS.

Code: Pascal  [Select][+][-]

1.   P.OSes := [linux, solaris] + AllBSDOses; // Darwin was tested!    

Can you add Solaris support in 3.2.4?

[msohn]

AFAICS https://gitlab.com/freepascal.org/fpc/source/-/commit/0758aa1143a6498532b3d100ed15edd63ca878b6 has not made it into 3.2.4, correct?

That is an absolute show-stopper for me and will also make it impossible to use fpcupdeluxe on the next version of macOS (27?) which will no longer support Rosetta, i.e. running x86_64 code on Apple Silicon.

Could this please be considered for a back port?

[zeljko]

Have you tried to merge that commit into 3.2.4 source ? Tested it ?

[dbannon]

Wow, thats one serious commit. Just about every platform individually affected ? No description about what it does, or why or how to test it. Two and a half years ago ?

msohn, you might need some help to get going some other way. Why does this problem particularly affect fpcupdeluxe ? Probably others ?  I build my app for Mac Arm users by cross compiling from an Mac Intel box and make a AARCH64 binary, there seems to be lots of other Mac users here using AARCH64 binaries. Why are you locked into Rosetta model ?

Davo

[msohn]

Yes it is a serious commit. I've been using it together with https://gitlab.com/freepascal.org/fpc/source/-/commit/c89848a00f59cc2539c5ff3862a547218129f594 (which contains the description the other commit was missing!) on top of FPC 3.2.2 for years now.

The problem with fpcupdeluxe is the bootstrap compiler it brings with it - for aarch64 that is the original 3.2.2 having the bug, so any FPC you're trying to build will crash - rendering the fpcupdeluxe useless, so one has to use the x86_64 version and build a cross compiler for aarch64 with it. That is what everyone seems to be doing until now, but that won't work anymore if you're on an OS that no longer has Rosetta. See also https://github.com/LongDirtyAnimAlf/fpcupdeluxe/issues/754, that is just one of many issues that users have been reporting there.

I just successfully made this work on the fixes_3_2 branch and verified that the fix is indeed working on aarch64. I cherry picked the first commit (0758aa11) and resolved all conflicts by choosing the file from that commit as is. I discarded all new files in that commit as I presumed these have been removed platforms (that also applies to the commit c89848 linked above).

I'll gladly provide a patch or open an issue on GitLab if that helps getting these fixes into 3.2.4!

[Graeme]

Oh, come on Graham, its over four years since we had an official release for ANY Platform !

Time to live dangerously and build FPC Trunk*.  🤣


* https://gitlab.com/freepascal.org/fpc/source/

[ALLIGATOR]

I may soon be banned from this forum
But, as always, I'll add my two cents

There's nothing wrong with FPC [main] 

[dbannon]

... so one has to use the x86_64 version and build a cross compiler for aarch64 with it.

OK, so if that works it would also be possible to cross compile from an Intel Mac to a aarch64 FPC binary ? I am thinking of a "work around" until a suitable FPC fix happens.

I have a system, for my one use, that allows me to put an up-to-date (ie rc1) FPC on a system without compiling it there. My 'library' of precompiled FPC already already includes an Intel Mac one. Separately, I already cross compile my app to aarch64, sounds like I could, easily, add a aarch64 FPC rc1 to that 'library' ?

https://github.com/davidbannon/FPC_Laz_Install - The install script has not been extensively tested on the Mac as I only have one, got it worked and stopped. I don't have a Mac aarch64 machine. But if you would like to experiment ...

I just successfully made this work on the fixes_3_2 branch and verified that the fix is indeed working on aarch64.

The issue there (I suspect, don't know) is testing on all the affected platforms. You have obviously tested on aarch64, how about making the patched source available so others can test on other platforms. I can do Linux [i386, x86_64, armhf, aarch64] maybe expressif. Who knows who else is listening in to this conversation ?

A clear description, or auto test, so we can be sure of the test results.

I guess a patch against the current 3.2.4rc1 ?   Or maybe the branch? No idea which the devs would prefer.

It might help, maybe we would be wasting our time but lets find out ?

Davo

[msohn]

Do you fully understand the broad effects this bug has? If you're building an app or dylib with aarch64 FPC 3.2.x and it contains any floating point math (hard to avoid, esp. if you link to other code, e.g. dylibs, frameworks etc.) and more than a single thread, you're bound for a hard crash. If you dig through it, https://gitlab.com/freepascal.org/fpc/source/-/issues/38230 has a lot of info from the commits author.

The bug is known for more than 5 years now, the fix is in trunk for 3.5 years, so I would consider it sufficiently tested.

My concern is that eventually it will be difficult to find a Mac that can still run x86_64. So you would have to use a different platform, e.g. Linux or Windows to build the cross-compiler and then you'd need to come up with a way of "installing" it on the Mac, like you apparently did. Nothing impossible, but far from the convenience of fpcupdeluxe.

I still hope one of the devs would consider the fix for 3.2.4. If so, I can try to provide whatever is needed.

[PascalDragon]

There's nothing wrong with FPC [main] 

If it works for you, great, but that is not the case for everyone and it currently contains enough major issues that definitely don't make us recommend using main for productive use compared to 3.2.2.