2008/7/14, Daniel Stenberg wrote:

> On Mon, 14 Jul 2008, Yang Tse wrote:
>
> > Finally here's the 64bit curl_off_t patch against current CVS.
>
> [...] I'm currently in vacation mode so it may take me a week
> or two to digest and get back with feedback on this.

That's cool. Have a great time!

I'm writing for everyone.

I really wanted to post it, so that others could start playing with
it, testing it, as soon as possible, to sort out unforeseen problems.
Or to get confirmation of working platforms.

I have already discovered a couple of issues:

First one, there was a minor problem with the include path of
libcurl's external headers include path for the libtest programs. This
is now fixed in CVS. The patch still applies with this change. and
should be tested with current CVS, so if someone is trying to test it
better make a new checkout.

On my system the patch does not apply cleanly for buildconf.bat. This
is a stupid CRLF thing which can be ignored if tests are going to be
done on a configure-capable platform. But if it is going to be tested
on a non-configure platform that hunk (2 lines) have to be merged
manually.

And now the nasty one which does mean more work. On configure-capable
systems it fails to configure/build out of the source tree :-(

This is due to the following: The generated curlbuild.h file is placed
in $(top_builddir)/include/curl/curlbuild.h and all inclusions are
expecting it in $(top_srcdir)/include/curl/curlbuild.h while the
library isn't installed.

While fixing this there is at least one thing which should not be
changed and it is the way in which curlbuild.h and curlrules.h are
included from curl/curl.h, otherwise non-configure builds would break.

The quick hack would be to copy
$(top_builddir)/include/curl/curlbuild.h to
$(top_srcdir)/include/curl/curlbuild.h at the end of the configure
script. But this would disable the possibility of doing parallel
builds with only one source tree.

I guess that the only way to go is modification of include paths for
the configure build process. But I would be glad to hear other
suggestions.

Did I say that it can be fully tested when building in the source tree ?

Ah! yes.

-- 
-=[Yang]=-