@fniephaus approved this pull request.

LGTM, don't know about TERF. Thanks, @nicolas-cellier-aka-nice!

On Mon, Apr 26, 2021 at 11:54:16AM -0700, Nicolas Cellier wrote:

This enables the linux builds again, while preserving windows and macos builds. That is a fix for https://github.com/OpenSmalltalk/opensmalltalk-vm/issues/560

If this is compatible with TERF VM, then it should be merged ASAP, the builds are red for too long. (we'll have to inquire next point of failure...)

You can view, comment on, or merge this pull request online at:

https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/562

Builds for the opensmalltalk-vm for Unix/Linux have been broken since February 8, 2001. The commit that caused the initial breakage is:

commit aafcb78371c7e576073a8dbf2f1f32667568e05e Author: Eliot Miranda eliot.miranda@gmail.com Date: Mon Feb 8 20:49:07 2021 -0800

All subsequent versions remain broken, and it has not been possible to build a VM for Linux since that time.

The commit notice includes the following request:

I ask anyone wishing to alter the system here to please contact me and discuss before making changes. It is extraordinarily expensive and frustrating to have to make changes of this kind to make sure that a signfiicantly more complex VM such as Terf's builds correctly.

In April 2021, Nicolas worked through the issues necessary to resolve this problem. He provided platforms and src fixes on GitHub in the fix_include_order branch, along with VMMaker.oscog-nice.2957 in the VMMaker inbox to provide the necessary code generation fix.

Technically, what we need next is: 1) Merge Nicolas' VMMaker fix into VMMaker.oscog, regenerate sources, commit to GitHub 2) Merge the fix_include_order branch into Cog

But we would also like to honor Eliot's request to review such changes before committing.

Eliot, could you please take a look at this and comment?

References: http://source.squeak.org/VMMakerInbox/VMMaker.oscog-nice.2957.mcz https://github.com/OpenSmalltalk/opensmalltalk-vm/tree/fix_include_order https://github.com/OpenSmalltalk/opensmalltalk-vm/pull/562 https://github.com/OpenSmalltalk/opensmalltalk-vm/issues/560

Thanks, Dave

We definitely need this asap.

So, I updated this PR to HEAD. I will merge this soon. Just testing MSVC builds now.

Some remaining quirks (using VMMaker.oscog-mt.2986 in VMMakerInbox, based in -nice.2957, also in inbox) are: - "#include <asm/cachectl.h>" moves to the bottom in cogitMIPSEL.c - Some hiccups (and extra type casts) regarding usqInt vs sqInt ... ? Maybe it is nothing, maybe there is some extra magic here during source generation that I am not aware of.

So, besides that include-order thingy in the VMMaker, there are only those extra HAVE_CONFIG_H in cross-platform sources: platforms/Cross/plugins/B3DAcceleratorPlugin/sqOpenGLRenderer.c platforms/Cross/plugins/IA32ABI/xabicc.c platforms/Cross/plugins/SoundCodecPrims/sqSoundCodecPluginBasicPrims.c platforms/Cross/plugins/Squeak3D/b3dInit.c platforms/Cross/plugins/Squeak3D/b3dMain.c platforms/Cross/plugins/SurfacePlugin/SurfacePlugin.c platforms/Cross/vm/sqHeapMap.c platforms/Cross/vm/sqVirtualMachine.c

Platform-specific fixes considering HAVE_CONFIG_H: platforms/unix/plugins/FileAttributesPlugin/faSupport.c platforms/unix/vm/aio.c platforms/unix/vm/sqUnixVMProfile.c platforms/win32/vm/sqPlatformSpecific.h

And this: platforms/unix/config/getversion.c platforms/Cross/vm/sqSCCSVersion.h platforms/Cross/vm/sqAssert.h

Note that sources have to be generated even if a dialog claims that there are not changes. Apparently, the order of include files is not considered a change in the generated artifact:

```Smalltalk [VMMaker generateAllConfigurationsUnderVersionControl] valueSupplyingAnswer: true. "Not false!" ```

Also note that the current VMMaker changes are in the VMMakerInbox: - VMMaker.oscog-mt.2986 based on - VMMaker.oscog-dtl.2968 based on - VMMaker.oscog-nice.2957

@eliotmiranda approved this pull request.

Looks fine to me. Clearly we need to regenerate sources. Deep apologies for having dragged my feet on this. A long dark night of the soul, now happily in the past (echo cancellation).

Thanks a bunch :D

Build order still a problem on build.linux32ARMv6/squeak.cog.spur/build [Tested on Raspberry Pi 3 on RasPi Linux]. Still have to comment out #ifdef _FEATURES_H in "config.h" as workaround

So, the build.linux32ARMv6 config does include "features.h" before "config.h" but that's not an issue for ARMv6? How come?

The only unix-specific changes were in these files: - platforms/unix/plugins/FileAttributesPlugin/faSupport.c - platforms/unix/vm/aio.c - platforms/unix/vm/sqUnixVMProfile.c

Don't those apply to the ARMv6 build as well? Would you post the details where "features.h" is included first?

Did a mvm build on RasPi3: build.kinux32ArmV6/squeak.cog.spur/build.

Changed "mvm" to add flag "-DUSEEVDEV", after which vm-display-fbdev built and VM runnable.

The -DUSEEVDEV flag should be set based on current pull request for fixes to vm-display-fbdev.

Looks good now.

Thanks much!