Show up for logs how we apply patches when we install ocaml-solo5 via opam

[dinosaure]

This patch does not change many things but it seems to fix the installation of ocaml-solo5 via opam into an ubuntu-latest GitHub machine. For the record, without this patch, I get an error (you can see here) and with this patch, it seems that we are able to install correctly ocaml-solo5 (you can see the output here - the error in the CI is unrelated to ocaml-solo5).

I suspect the set -x but it's highly related to #151. @shym do you have any insights?

[dinosaure]

So we fallback to what we did about esperanto (for the context, see dinosaure/esperanto#52) and we must call git init -q. By this way, ocaml-solo5 is able to be compiled by dune pkg.

[reynir]

Just wondering: do all patches apply no matter the ordering? With patches/$VERSION/* I think we are not guaranteed any particular order, and it will depend on the filesystem. I see the patches go from 0001-*.patch..0021-*.patch.

[reynir]

Yes, I think we need to force the order. I tested with the patches in reverse order and it fails:

$ git apply --check --directory ocaml patches/5.3.0/0021-Set-Max_domains_-def-max-to-1.patch patches/5.3.0/0020-Allow-ocaml-in-a-triplet-target-to-build-freestandin.patch patches/5.3.0/0019-Accept-native-freestanding-targets-at-configure-time.patch patches/5.3.0/0018-Output-the-.size-and-.type-directives-based-on-confi.patch patches/5.3.0/0017-Add-a-Config-variable-for-the-.size-and-.type-assemb.patch patches/5.3.0/0016-Detect-support-for-.size-and-.type-assembler-directi.patch patches/5.3.0/0015-Output-the-.note.GNU-stack-section-based-on-configur.patch patches/5.3.0/0014-Add-a-Config-variable-for-the-need-of-the-GNU-non-ex.patch patches/5.3.0/0013-Detect-the-need-for-the-GNU-note-for-non-executable-.patch patches/5.3.0/0012-Include-asm.h-at-the-beginning-of-assembler-sources.patch patches/5.3.0/0011-Add-a-dummy-caml_debugger_saved_instruction-when-HAS.patch patches/5.3.0/0010-Fix-static-builds-of-the-compiler.patch patches/5.3.0/0009-Enable-bootstrapping-flexdll-in-the-cross-compiler-s.patch patches/5.3.0/0008-Add-a-Makefile.cross-with-rules-to-build-a-cross-com.patch patches/5.3.0/0007-Detect-flexlink-only-on-relevant-targets.patch patches/5.3.0/0006-Add-a-configurable-library-directory-on-target.patch patches/5.3.0/0005-Use-a-TARGET_BINDIR-configure-variable-instead-of-wi.patch patches/5.3.0/0004-Check-that-the-OCaml-versions-are-compatible-for-a-c.patch patches/5.3.0/0003-Detect-a-_build_-C-toolchain-to-build-sak.patch patches/5.3.0/0002-Use-target-instead-of-host-when-relevant-in-configur.patch patches/5.3.0/0001-Use-target-instead-of-host-to-detect-the-C-toolchain.patch
error: the patch applies to 'ocaml/configure' (401b94298e4f34946677a409560ce1832c5bbbbc), which does not match the current contents.
error: ocaml/configure: patch does not apply
error: patch failed: ocaml/configure.ac:299
error: ocaml/configure.ac: patch does not apply

[shym]

Indeed the order must be enforced. I wrote it that way because IIUC sh is expected (by POSIX) to sort entries when expanding wildcards. Now I discover bash parameter $GLOBSORT that can be used to control (including disabling) how sorting is done, so this is another brittleness of that patching code. Sigh.
Anyway, in the log I don’t see any failure due to some patch not applying, so I don’t think that the order of the patches is causing issues.

[reynir]

FWIW I got the same error message when I did git apply --check directory ocaml patches/5.3.0/* and adding echo in front it's in the numerical order. So I don't know what's going on and why I'm getting that error.

[dinosaure]

I still think that using git apply is not appropriate because it also looks at .git/index. It is much more than just a patch between two files; it is a patch between two files and the state of the Git repository we are in (and it is difficult to know how git apply finds .git).

In short, I think we're back to the initial issue described here: #151

[shym]

This patch does not change many things but it seems to fix the installation of ocaml-solo5 via opam into an ubuntu-latest GitHub machine. For the record, without this patch, I get an error (you can see here) and with this patch, it seems that we are able to install correctly ocaml-solo5 (you can see the output here - the error in the CI is unrelated to ocaml-solo5).

I suspect the set -x but it's highly related to #151. @shym do you have any insights?

As opam hides the successful log, it’s hard to compare the logs.
(I’m not sure what you have in mind about set -x there.)

So, what is strange is that git isn’t complaining that patches cannot be applied (I just checked that the error appears even without -v, fortunately, at least in my local version of git; but I don’t expect such errors to be hidden without -v anyway). And still, the error in the ./configure call is clearly signalling that configure wasn’t properly patched.

What comes to mind:

• We might turn the ; into a && between VERSION=... and for... to make sure the VERSION file was indeed read properly; and maybe log it explicitly? I don’t see how it could be the case there, but an empty VERSION could explain what we saw.
• I look forward to OCaml 5.4, where we could decide to get completely rid of patches: only configure and the default number of domains will need to be changed, maybe we should go there with a simple cp instead of trying to be clean.

[hannesm]

is this superseeded by #154?

[dinosaure]

Yes, thanks for the work.