Re: How about some AC_MSG for dlopen?

Andreas Metzler <[email protected]> writes:

> On 2024-08-19 Daiki Ueno <[email protected]> wrote:
>> Andreas Metzler <[email protected]> writes:
>> > On 2024-08-17 Daiki Ueno <[email protected]> wrote:
>> >> Andreas Metzler <[email protected]> writes:
>
>> >> > ./configure is quiet about whether dlopening librariers is supported,
>> >> > would it make sense to a) add messages and perhaps b) provide a switch
>> >> > to force use of linking instead?
> [...]
>> I would prefer to have the option per library, because some of the
>> libraries are less likely to be available than the others (e.g.,
>> tpm2-tss vs. zlib).
> [...]
>
> Per-library quadstate ends up being rather wordy, see attached example
> for handling zlib that way. I have thought about it but cannot see a
> substantial abbreviation potential ad_hoc, there are 8 cases to handle.
> - Do you want me follow-up, extending this approach to brotli/zstd/tpm?

Thanks.  A couple of comments:

- It would be simpler to use AC_MSG_CHECKING/AC_MSG_RESULT for
  ENABLE_DLOPEN rather than AC_MSG_NOTICE
- Perhaps we could drop the non-pkgconfig detection logic for zlib, as
  the library supports pkgconfig since August 2006
- I would first normalize the argument for --with-*, something like:

  AC_ARG_WITH(zstd,
              AS_HELP_STRING([--without-zstd], [disable zstd compression support]),
              with_zstd=$withval, with_zstd=yes)

  AS_CASE([$with_zstd],
          [yes], [AM_COND_IF([ENABLE_DLOPEN], [with_zstd=dlopen], [with_zstd=link])],
          [dlopen], [AM_COND_IF([ENABLE_DLOPEN], [:], [AC_MSG_ERROR([[
  ***
  *** Unable to dlopen ZSTD, try --with-zstd=link.
  *** ]])])],
          [link], [:],
          [no], [:],
          [AC_MSG_ERROR([[Unknown argument $with_zstd for --with-zstd]])])

  So afterwards $with_zstd should only be dlopen, link, or no.

- Perhaps we might want to provide a autoconf macro or two for common
  tasks, in m4/hooks.m4, to avoid code duplication

Regards,
-- 
Daiki Ueno