Re: setup.exe-HEAD should build from CVS on gcc-3.

["Max Bowsher" <maxb at ukf dot net>]

Hmm... seems I've discovered a new kwyboard shortcut. Ctrl+Enter = Send.

So my previous message is incomplete.

Corrected version below:
----------------------------------------------------------

Robert Collins <rbcollins@cygwin.com> wrote:

> Ok, I've applied most of Max's patches. CVS setup now builds and runs
> for me with gcc-3.

Me too :-)

> Max, I've not applied a couple of the patches:
> -Wno-deprecated is not needed.

Yep.

> proppage.h I've not gotten to (yet).

That problem has silently dissappeared... for now.
You see that I added #define NOMINMAX above the #include <windows.h> in
win32.h? Well these other places in the code #include <windows.h> directly:

cistring.h:18:#include <windows.h>
proppage.h:23:#include <windows.h>
propsheet.h:25:#include <windows.h>
window.h:23:#include <windows.h>
window.cc:20:#include <windows.h>

We scrape by, so long as none of these happen before a STL include, or, if
they happen before a STL include, win32.h was included first (causing a
no-op due to double-inclusion guards).

Possible actions:
1) Do nothing until this resurfaces, then reorder the relevant includes to
get rid of it.
2) Change the above to include win32.h instead.
3) Create and use a mywindows.h, which is simply #define NOMINMAX, #include
<windows.h>
4) Add #define NOMINMAX above all above.

> I don't think I've missed any other patches, can you diff your sandbox
> against my commits to see ?

No, all important patches accounted for.

There was one other, fixing support for Cygwin-linked compilation.
Comment in port.h:

Robert Collins <rbcollins@cygwin.com> wrote:

> Ok, I've applied most of Max's patches. CVS setup now builds and runs
> for me with gcc-3.

Me too :-)

> Max, I've not applied a couple of the patches:
> -Wno-deprecated is not needed.

Yep.

> proppage.h I've not gotten to (yet).

That problem has silently dissappeared... for now.
You see that I added #define NOMINMAX above the #include <windows.h> in
win32.h? Well these other places in the code #include <windows.h> directly:

cistring.h:18:#include <windows.h>
proppage.h:23:#include <windows.h>
propsheet.h:25:#include <windows.h>
window.h:23:#include <windows.h>
window.cc:20:#include <windows.h>

We scrape by, so long as none of these happen before a STL include, or, if
they happen before a STL include, win32.h was included first (causing a
no-op due to double-inclusion guards).

Possible actions:
1) Do nothing until this resurfaces, then reorder the relevant includes to
get rid of it.
2) Change the above to include win32.h instead.
3) Create and use a mywindows.h, which is simply #define NOMINMAX, #include
<windows.h>
4) Add #define NOMINMAX above all above.

> I don't think I've missed any other patches, can you diff your sandbox
> against my commits to see ?

No, all important patches accounted for.

There was one other, fixing support for Cygwin-linked compilation
(reattached)
Comment in port.h:

/* I prefer building a cygwin version of setup for debugging, as I
   find that gdb can debug those programs (and the exceptions they
   cause) better.  This file handles the slight differences between
   cygwin and mingw. */

I assume based on the copyright bit, this was written by D J Delorie.

Anyway, this mode is currently broken. The patch fixes it up - in a slightly
kludgy way.

> Also, I've remembered the bug. HEAD selects the wrong default package
> (i.e. running it here it defaulted to openssl ...9.6e.) AFAIK that's
> the only bug preventing release.

Yes, this is the one I've seen.

> HEAD has debian Releases file
> support and many performance improvements, and more user feedback, so
> I would like to get it out. Until I've moved though, I won't have the
> time to do in-depth analysis. I was lucky to squeeze in the setup
> update as it is.

Sure - no problem. I have a couple of improvements, (Categories column,
Colour code prev/curr/exp) pending, so let me know when the chaos subsides.

Max.