[m-rev.] for post-commit review: canonicalizing path chars

[Julien Fischer]

Hi Zoltan,

On Fri, 8 Sep 2023, Zoltan Somogyi wrote:

> By the way, the bootcheck for this diff got me an error I haven't seen
> for a while:  "mismatch in list of stdlib modules".  The reason seems to be
> that both DOC_VIA_MMC0 and UNDOC_VIA_MMC0, the files from which
> one of the two lists to be compared is obtained, both contain
>
> mercury_compile:
> unrecognized option `--fullarch`
>
> The strange thing is that both these files are produced by invoking
> "mmc --output-stdlib-modules", and invoking that command manually
> gives me the expected file list, not the above error. Also, a grep for
> "fullarch" on all the files under the top directory of my workspace
> did not result in a single match on "fullarch" being used as an option.
>
> Has anyone else seen this, and does anyone have ideas on
> what may be going on here?

Yes, I saw it yesterday.  It is (obviously) due to my deletion of the
--fullarch option. I *think* what is happening is that when stage 2 runs
the check_stdlib_modules target, the bootcheck script hassets
MERCURY_COMPILER to point to the stage 2 compiler, but it still looks at
the bootstrap compiler Mercury.config (for some reason.)

The immediate problem resolves itself when the bootstrap compiler has
the change that deleted --fullarch installed, although it now raises the
question of: is stage 2 (always) actually looking the correct .config
file?

Julien.