Re: [Boost-users] Sunpro CC 5.8, -library=stlport4, multi_index, one last time

From: Joaquín Mª López Muñoz (joaquin_at_[hidden])
Date: 2007-06-25 09:22:29

Hi Benson,

Benson Margulies ha escrito:

> I beg the indulgence of the maintainers to clarify the support
> intentions here and then I'll go back to my cave.

Please stay out of the cave :) I'm afraid there's no appointed group
of people to whom you can officially refer as "the maintainers"
--Boost structure is rather informal, some folks care about some
particular libs, others contribute regression testings for their platforms
of interest, etc. So, supporting a particular lib in a particular platform
necessitates a willing author/contributor and a willing tester --maybe
you can be the willing tester. Without a stable regression tester
for a particular platform, it's extremely unlikely that that platform can
receive proper attention.

> The facts are:
>
> 1) The regression tests run reasonably on CC 5.8, with or without
> -library=stlport4

Regression tests are satisfactory for -library=stlport4,

http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/developer/summary.html

(column sun-5.8)

but there is no current regression tester running without stlport4, so we
can't know how well that option would perform. The -library=stlport4
tester is Doug Gregor,

http://engineering.meta-comm.com/boost-regression/CVS-RC_1_34_0/OSL4-V2.html

so maybe he can tell us why he opted for stlport4 and what the general
behavior without that is.

> 2) I have a seemingly simple example of using multi_index that won't
> compile with CC 5.8 unless I'm using -library=stlport4. The compile
> error, for what it's worth, does not point to one of the weaknesses of
> the 'old Sun STL' in any obvious way. But it could easily be that some
> ifdef that is trying to compensate for the old STL is in turn creating
> the conditions for the failure.

I am as mystified as you wrt to this particular error :( Having a test run
for all Boost libs would certainly give us a more comprehensive vista on
the general behavior of CC 5.8 without stlport4 --and could help us
determine whether this is a fixable thing or if we're beyond all hope
of supporting the compiler.

> (I'm not entirely sure about #1.)
>
> The question is:
>
> "Is it the intention of the maintainers to support CC 5.8 without
> -library=stlport4?"
>
> If the answer is 'yes', then I will nudge to get this case included in
> the regression, and I will invest some time in trying to unscrew the
> inscrutable error.
>
> If the answer is 'no', then I will concentrate on moving our Apatosaurus
> of a code-base to getting itself compiled with -library=stlport4 as soon
> as I can, and defer use of boost until that can accomplish itself.

If there is a stable regression tester running CC 5.8 without stlport4,
I'll certainly take a look at the failures in Boost.MultiIndex and try
to fix them. Maybe you're in a position to be that tester. Without testing
support I'm afraid my help would be much less effective, since the platform
is not so easily available, although of course you can count on my
assisting you nevertheless.

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo