Build failures on OS X 10.3.9: isnan, isinf, etc.
Hello, I've been looking at building source from SVN on OS X 10.3.9 and I've found these problems:
- configure.ac uses g++ instead of gcc when testing for -Wno-pointer-sign support (ignoring $CFLAGS) and gcc3 dies in libcroco. Already reported by jiho.
- isnan undeclared in src/libcola/cola.cpp. I worked around this by changing "isnan" to "isNaN" since the latter is defined in isnan.h, though I'm not sure this makes sense for all platforms.
- libcola/gradient_projection.cpp:54 isinf undefined. Adding '#include <math.h>' fixes this.
There are several issues for isnan already in the tracker, so my question is: are these new problems (requiring new bugs created), or the same as earlier bugs?
- Although configure apparently finds msgfmt, In po/Makefile, both $MSGFMT and $GMSGFMT are empty. When make hits the '.po.gmo:' target, it complains "-o: command not found" since part of the command is "$(GMSGFMT) -o $$file $<". Has anyone run into this?
Thanks!, John
It seems to me that OS X 10.3.9 gives a really disproportionate amount of problems. We have dozens of bugs filed for it. Just curious, how common is it? Can we leave it behind as we did for Windows 98? If not now, when will it be painless enough?
On 2007-April-19 , at 17:54 , bulia byak wrote:
It seems to me that OS X 10.3.9 gives a really disproportionate amount of problems. We have dozens of bugs filed for it. Just curious, how common is it? Can we leave it behind as we did for Windows 98? If not now, when will it be painless enough?
It was already kind of "left behind" because we could not get a 0.45 package done for it. Concerning the bugs in the tracker, most of them refer to the same problem just in different incarnations. I took some time to review them but not to merge the comments/recommendations and close them yet. But actually, at least _some_ of the issues John reports are more related to gcc 3.3 than to OS X 10.3.9 itself: I have the same when trying to compile Inkscape with gcc 3.3 tough I use OS X 10.4. I just think it is better for him to report them that for me because his system is entirely gcc 3.3 based while using gcc 3.3 on mine leads to an uncommon mixture of gcc 3 and gcc 4 which could lead to un- necessary bugs. A better solution to isolate gcc 3 related issues from OS 10.3.9 related issues would have been to have two people with the same OS version and different gcc versions. It was our initial plan but, first, I haven't heard back from the other 10.3.9 folks and, second, John could not get gcc 4 to compile on his system, so here we are.
- configure.ac uses g++ instead of gcc when testing for
-Wno-pointer-sign support (ignoring $CFLAGS) and gcc3 dies in libcroco. Already reported by jiho.
This is surely a gcc 3 problem. John provided a fix for it. I asked for a review of his patch on the list because I am not qualified enough to review it myself. No feedback yet.
- isnan undeclared in src/libcola/cola.cpp. I worked around this by
changing "isnan" to "isNaN" since the latter is defined in isnan.h, though I'm not sure this makes sense for all platforms.
- libcola/gradient_projection.cpp:54 isinf undefined. Adding
'#include <math.h>' fixes this.
There were some OS 10.3.9 specific isnan, isnormal, .... problems reported before. Inkscape used to compile on 10.3.9 so Michael re- established the OS 10.3 specific definitions in isnan.h, isnormal.h and these problems were gone. These new problems John hits seem related to recent changes to the codebase (i.e. posterior to the last successful compile of Inkscape on 10.3.9 with gcc 3) so I don't know to which extend they are gcc 3 or OS 10.3.9 related. Anyway, isnan.h, isnormal.h and friends seem to be a problem for all cross-platform work. Bob Jamison and other came up with suggestions in my previous gcc 4 vs. gcc 3 thread: http://sourceforge.net/mailarchive/message.php? msg_id=AF88F7EC-2202-4EEC-ADFC-8A89C0F41A20%40gmail.com what about them?
JiHO --- http://jo.irisson.free.fr/
participants (3)
-
bulia byak -
jiho -
John Faith