instr_size fixes for arm64

[tmcgilchrist]

This is a rebase of the work originally done in #11162 by @stedolan

The main change is adding a configure flag --enable-debug-codegen to turn on the debugging assertions on instr_size so it is opt in (285db26), and I have added this configure flag to GitHub CI so it will run in CI (d234440)

@xavierleroy I think that last point covers what you suggested here #11162 (comment)

This should address the bug report #13596

[dra27]

@xavierleroy - the check is now restricted to GitHub Actions (and can be generally enabled via a new configure option). Does that adequately deal with your concerns from earlier?

[xavierleroy]

Looks good to me.

One very minor reservation: the name debug_codegen (in Config) and the option --enable-debug-codegen suggest that this is about generating debugging information (a.k.a. "GDB support"), in which case users will be disappointed. Maybe --enable-codegen-assertions ?

[tmcgilchrist]

That is a good point. I'm ok with --enable-codegen-assertions. @dra27 Are you happy with the suggested naming ?

@dustanddreams Did a similar tidy up of the instruction size calculations for Power #13221 that I would like to re-use this flag for. I'll submit that in a follow up PR.

[dra27]

Name works for me - possibly also --enable-codegen-invariants to match --enable-flambda-invariants and --enable-cmm-invariants?

[xavierleroy]

possibly also --enable-codegen-invariants

Fine with me. Pick whichever you prefer.

[dra27]

Running through precheck - aside from the one line of YAML that needs deleting, this is good to go, assuming precheck doesn't through up anything.

[dra27]

Suggested change

config_arg: CONFIG_ARG='--enable-codegen-invariants'

[gasche]

Merged, thanks!