2001-07-15 Michael Natterer <mitch@gimp.org>
* libgimpbase/gimpwire.c: remove the usage of printf()/scanf()
when sending doubles over the wire. Instead, rely on the memory
layout of gdouble being IEEE compliant and transmit 8 bytes in
network byte order.
* libgimpbase/gimpprotocol.h: increase GP_VERSION because this
makes the wire protocol binary incompatible.
* app/main.c
* libgimp/gimp.c: removed the setlocate(LC_NUMERIC,"C") workaround.
2000-05-31 Michael Natterer <mitch@gimp.org>
* app/gimpunit.c
* libgimp/*: all libgimp headers are included via gimp.h or
gimpui.h, so include <gtk/gtk.h> there and in the *.c files.
Various cleanups.
2000-05-14 Michael Natterer <mitch@gimp.org>
After a weekend of extensive mail exchange with the signal crew,
we finally found that #2742 is not a Gimp bug but a real OSF/1
bug. OSF/1 does _not_ reliably restart read() and write() calls
on certain slow devices (pipes).
* libgimp/gimpwire.[ch]: guard all read/write calls on the wire
with loops checking for EINTR. This has to be done "manually"
as glib's GIOChannels return G_IO_ERROR_UNKNOWN on the occurence
of EINTR (which is a bug, too).
s/int/gboolean and minor cleanups while I was on it (not changing
any logic).
This fix depends on the current (broken) state of the GIOChannel
implementation and is scheduled for removal as soon as glib
behaves nicely here.
Left SA_RESTART there for the moment in app/main.c. See it as
defensive programming or just my fear to change two #2742-related
places at the same time. We might choose to remove SA_RESTART
later. Many thanks again to Austin, Garry and Tim.
2000-05-10 Michael Natterer <mitch@gimp.org>
Another try to get the signal/dead child recovery stuff right.
Could the brave signal crew (TM) (Austin, Garry, Raphael, Tim)
please do bad tests to the new code? I removed all strange
constants (SA_NODEFER etc.) and used only glib and POSIX stuff.
* app/main.c
* libgimp/gimp.c:
- Call gimp_signal_private() with no flags to enforce a proper
sigaction() behaviour (block signals while handler is active).
- Removed the reentrancy guards from the handlers.
- Renamed the handlers.
- Ignore SIGPIPE in the app and in plugins.
- Re-introduced the SIGCHLD handler because it should work
now. Also added a SIGCHLD handler to libgimp/gimp.c.
* app/errors.c
* libgimp/gimp.c: in the signal handler, unblock all signals
with sigprocmask() before calling g_on_error_query() because
gdb otherwise inherits the blocked signals and does nothing.
Wrapped the statements with "if (TRUE) { }" blocks so it's
easy to make the stack trace a command line option.
* app/plug_in.c
* libgimp/gimp.c: listen for G_IO_ERR and G_IO_HUP on the read
channels. In the app, pop up an error message and clean up the
plugin. In plugins, clean up and exit.
* libgimp/gimpwire.c: removed the "plug-in chrashed?" message
and print the program's name with all error messages.
* plug-ins/helpbrowser/helpbrowser.c: typo.
* app/app_procs.c
* app/gdisplay_color.[ch]
* app/gdisplay_color_ui.c: make the ui usable
* app/gdisplay.c: enable cdisplay support
* modules/Makefile.am
* modules/cdisplay_gamma.c: moved gamma functionality to separate
-Yosh
* libgimp/gimp.c: No need for <io.h> on Win32.
* libgimp/{gimp,gimpui}.def: Add two entry points.
* libgimp/gimp.h: Can't use __declspec(dllexport) for PLUG_IN_INFO
when compiling with gcc on Win32. Also handle __argc, __argv and
_stdcall differently with gcc on Win32
* libgimp/gimpenv.c: Include <string.h>.
* libgimp/gimpfeatures.h.in: Remove lots of extraneous trailing blanks.
* libgimp/gimpfileselection.c: Include <glib.h> early, see above.
* libgimp/{gimpwire,parasite}.c: Test for NATIVE_WIN32, not
_MSC_VER.
* libgimp/makefile.msc: Miscellaneous updates. The gimpi library
is now built as a static library.
Michael Natterer