Re: [Gc] Support for mingw + threads
ivmai at mail.ru
Fri Jul 24 12:12:02 PDT 2009
Juan Jose Garcia-Ripoll <juanjose.garciaripoll at googlemail.com> wrote:
> 2009/7/24 Ivan Maidanski <ivmai at mail.ru>:
> > Does ECL use unmodified version of GC v7.1?
> Mostly. The patch for signal handling in OS X is essential for ECL to
> build properly on that platform, but I believe you have committed that
Not. I've just sent it to the ML (so, at present, get v72a2+ and apply it yourself - should work for you).
> > If yes then you should switch to GC v7.2a2
> > (or if you are using incremental mode then better use CVS).
> I have problems accepting to move to an alpha release. Already too
> many complains from unhappy users about the move 6.8 -> 7.1 (needed
> for the OS X)
Among v6.8, v7.1[a...] and v7.2a2 - the versions 7.1 (alpha and "final") are the most bad. (So, it's not a surprise that the users complain about 6.8 to 7.1 switch.) Does anybody know the a thing that works in v7.1 but doesn't in v7.2a2? A lot of code polishing has been done since v7.1 and only a few new features have been implemented (like parallel collection on Win32 and threads support for SunOS x86 (the later is still not committed)).
I know that people typically afraid of "alpha" versions unless they know it deeply inside-out. But v7.2a2 should work better for you than v7.1.
> > It works with mingw gcc4+ (and with mingw-w64(32) of latest build if -DGC_NOT_DLL supplied);
> > -DUSE_MUNMAP, -DPARALLEL_MARK and -DTHREAD_LOCAL_ALLOC works too.
> I do not know about those flags. Are they documented anywhere?
Yes, see Makefile.direct.
> > Note 2: the building scripts may be not up-to-date.
> This does not sound like something I would like to distribute right
"does not sound like something I would like to distribute" or
"sounds like something I would not like to distribute" ? ;)
> now with ECL, specially when I have users that complain if they have
> to type "make clean" before configuring :-)
Again, the things (with the scripts) in v7.2a2 are better than in v7.1.
I only mean that some flags (like -DUSE_MUNMAP, -DPARALLEL_MARK and -DTHREAD_LOCAL_ALLOC) speeding up the collector could be placed (or auto-gen if --enable-parallel or so) in scripts now (since they are now working for Win32) - but the collector works even w/o them (as before).
It would be nice if someone improves the things...
> In any case, can I just prompt corageous users to build 7.2a2
> themselves if they want thread support + mingw32?
Feedback and suggestions/fixes are welcome...
> Instituto de FМsica Fundamental, CSIC
> c/ Serrano, 113b, Madrid 28006 (Spain)
More information about the Gc