On 2024-11-14 17:42, Divya Ranjan wrote:

 Nicolas Graves <ngraves@ngraves.fr> writes:

 On 2024-11-13 04:47, Divya Ranjan via Guix-patches via wrote:

 Hello Nicholas, Lars and others.

 I’ve planned to pick up the work needed for this upgrade. Where are we
 and what more work is needed? A brief summary with specific tasks
 would help me get started.

 I haven't managed to get much more done.  What happened is that
 core-updates broke the original patch series, the patch I added fixes
 the build of version 9.4.8 and makes a previous patch unnecessary, but
 the way I wrote it required to build from versions 8.6 (basically one
 entire day of pure build on my machine).

 If I were to rebuild everything from 8.6 once again, I would actually
 rather try the #:make-flag EXTRA_HC_OPTS (IIRC) which is definitely the
 same thing in 9.4.8, but it wasn't that clear in 8.6.

 Is building everything from 8.6 a good idea though? Is it /that/
 broken?

The upside is more consistent and easier to understand code in Guix, the
downside is rebuild times.  Only depends on how you weight each other.

From there you'll see that some tests for 9.6 are still broken.  I last
was working on decoupling the build (which works) from the tests (some
still failed, hard to understand why), because the rebuild is huge and
makes iterations quite painful.  But even that is hard since you would
need the hadrian test phase to be run to get the necessary files
(hadrian config for tests and some binaries) to run tests independently
in another guix package.  IIRC I stopped there but still have some
progress in my stash. 

 Okay, is there any particular reason why this is being so hard? I
 haven’t seen such problems with Nix.

It is hard to get what's wrong based on the error messages.  Also when I
tried to add options for more logging (don't remember which one I
tested) on Hadrian's options passed to the linker (through
EXTRA_HC_OPTS for instance), it fails because the option was
incompatible with another one set by Hadrian.  I wanted to try more
logging options to at least have a better understanding of what was
going wrong, but then started on decoupling build and tests because of
the high rebuild time.

There's still ultimately the option to skip the tests, which is not
difficult, but neither desirable. 

 Regards,