public inbox for insight@sourceware.org
 help / color / mirror / Atom feed
* Build for mingw32 and i686 on x86_64 observations
@ 2010-03-24 20:21 Gene Smith
  2010-03-31  5:26 ` Keith Seitz
  0 siblings, 1 reply; 2+ messages in thread
From: Gene Smith @ 2010-03-24 20:21 UTC (permalink / raw)
  To: insight

First off, I want to point out that I can build the very recent cvs head 
for an embedded arm application but it doesn't run correctly in that the 
insight gui does not reflect the actual location of the PC while 
debugging/stepping (the green highlighted line never moves). This is 
regarless of whether the tk/tcl is system supplied or insight's own.

Insight 6.8 does not have this problem. So I tried to build 6.8.1 to see 
how it does. With 6.8 there were kludges I had to do to get it to build 
and now those don't seem to be required when built on x86_64 (and 
probably i686). I also did a cross build on x86_64 for i686/32-bit and 
it worked with some effort:
additional configure options:
--host=i686-pc-linux-gnu --build=i686-pc-linux-gnu \
    CFLAGS="-m32" LDFLAGS="-m32"
and in files tcl(and tk)/unix/configure had to add $CFLAGS like this:
CC="$CC -pipe $CFLAGS"
to fix a build error.

But when build with i686-pc-mingw32-gcc toolchain (fedora 12 yum), the 
same (windows specific) kludges as before were required (syntax errors 
in window specific tck/tk code regarding dde and registry that can be 
commented out).

However, with mingw32 I had to build/install then build/install again to 
get insight.exe to appear at install/bin. It seems that in the install 
directory under lib there needs to exist *at compile time* tkConfig.sh 
and tclConfig.sh. So you have to make clean all, make install, then make 
clean all, make install again when you are starting with an empty 
install directory. So if you keep install/lib/tclConfig.sh and 
tkConfig.sh between compiles (don't completely clean the install dir) 
you are ok the next time.

-gene




^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Build for mingw32 and i686 on x86_64 observations
  2010-03-24 20:21 Build for mingw32 and i686 on x86_64 observations Gene Smith
@ 2010-03-31  5:26 ` Keith Seitz
  0 siblings, 0 replies; 2+ messages in thread
From: Keith Seitz @ 2010-03-31  5:26 UTC (permalink / raw)
  To: Gene Smith; +Cc: insight

On 03/22/2010 07:47 PM, Gene Smith wrote:
> First off, I want to point out that I can build the very recent cvs head
> for an embedded arm application but it doesn't run correctly in that the
> insight gui does not reflect the actual location of the PC while
> debugging/stepping (the green highlighted line never moves). This is
> regarless of whether the tk/tcl is system supplied or insight's own.

What target did you build for? Your configure options only list --build 
and --host.

Also, what target are you trying to debug under? The problem you 
describe is typically a bug in the target back-end. It would not 
surprise me to find out that something changed in gdb-land that broke 
insight. This happens more often than I would like, but since I only use 
insight natively, unless I can reproduce on a simulator, my bug-solving 
is limited to what I can divine by reading the code.

> But when build with i686-pc-mingw32-gcc toolchain (fedora 12 yum), the
> same (windows specific) kludges as before were required (syntax errors
> in window specific tck/tk code regarding dde and registry that can be
> commented out).

That doesn't surprise me. Our tcl/tk are greatly out of date, and I do 
not have general access to a windows machine. I've never tried building 
a cross from linux, though...

> However, with mingw32 I had to build/install then build/install again to
> get insight.exe to appear at install/bin. It seems that in the install
> directory under lib there needs to exist *at compile time* tkConfig.sh
> and tclConfig.sh. So you have to make clean all, make install, then make
> clean all, make install again when you are starting with an empty
> install directory. So if you keep install/lib/tclConfig.sh and
> tkConfig.sh between compiles (don't completely clean the install dir)
> you are ok the next time.

Configure will look for t{cl,k}Config.sh in the install directory and 
several other places. Wherever it finds those files, the location will 
be hard-coded into configure's cache (if what I am deducing from my 
vague recollection of how this all works is accurate). Your description 
makes sense. That is simply a quirk of how configure works.

Next time, try erasing the gdb directory and forcing a reconfigure.

Keith

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2010-03-31  4:28 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-03-24 20:21 Build for mingw32 and i686 on x86_64 observations Gene Smith
2010-03-31  5:26 ` Keith Seitz

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).