| msg18270 - (view) |
Author: Richard Jones (richard) *  |
Date: 2003-09-23 07:50 |
| Anthony said to log this: checking ncurses.h presence... yes configure: WARNING: ncurses.h: present but cannot be compiled configure: WARNING: ncurses.h: check for missing prerequisite headers? configure: WARNING: ncurses.h: proceeding with the preprocessor's result configure: WARNING: ## ------------------------------------ ## configure: WARNING: ## Report this to bug- autoconf@gnu.org. ## configure: WARNING: ## ------------------------------------ ## I'll attach the text of the report and the configure output too since I'm pretty sure the above won't be ligible once sf's tracker's finished with it. |
|
|
| msg18271 - (view) |
Author: Anthony Baxter (anthonybaxter)  |
Date: 2003-09-23 08:14 |
| Logged In: YES user_id=29957 creosote (which is a freebsd machine) also sees this. |
|
|
| msg18272 - (view) |
Author: Richard Jones (richard) * |
Date: 2003-09-23 08:28 |
| Logged In: YES user_id=6405 "make test" also reports: test_curses test test_curses crashed -- _curses.error: curs_set() returned ERR |
|
|
| msg18273 - (view) |
Author: Martin v. Löwis (loewis) * |
Date: 2003-09-27 07:53 |
| Logged In: YES user_id=21627 Can you try the patch in python.org/sf/806800 ? |
|
|
| msg18274 - (view) |
Author: Richard Jones (richard) * |
Date: 2003-09-27 08:36 |
| Logged In: YES user_id=6405 (I didn't mention it in the report, only to Anthony : this is on OSX) The patch doesn't alter the configure behaviour. I can't remember how to run the curses test - "make test" skips it (the curses module is built though). |
|
|
| msg18275 - (view) |
Author: Anthony Baxter (anthonybaxter) |
Date: 2003-09-30 14:24 |
| Logged In: YES user_id=29957 Looking into it a bit on FreeBSD: /usr/include/ncurses.h:236: conflicting types for `wchar_t' /usr/include/stdlib.h:58: previous declaration of `wchar_t' and poking around a bit I find that stdlib.h defines wchar_t as an 'int', while ncurses.h defines it as 'unsigned long'. This is on FreeBSD 4.8. Extracting the failing configure code and making it not define _XOPEN_EXTENDED_SOURCE makes the test succeed, but I have no idea if this is an appropriate fix. |
|
|
| msg18276 - (view) |
Author: Richard Jones (richard) * |
Date: 2003-09-30 23:29 |
| Logged In: YES user_id=6405 The diagnosis on OS X is similar, and the trigger for defining wchar_t in stdlib.h is the absence of _ANSI_SOURCE. Making it not define _XOPEN_SOURCE_EXTENDED also fixes the compile as per your FreeBSD fix. |
|
|
| msg18277 - (view) |
Author: Anthony Baxter (anthonybaxter) |
Date: 2003-10-01 08:52 |
| Logged In: YES user_id=29957 Unless a FreeBSD or Mac OS X expert can step forward in the next day and suggest a fix, this isn't going to be fixed for 2.3.2(final). I can't spend any more time on it. |
|
|
| msg18278 - (view) |
Author: Jack Jansen (jackjansen) * |
Date: 2003-10-01 15:04 |
| Logged In: YES user_id=45365 I've also seen this. I don't have the time to dive into this, but I think it has to do with gcc 3.3. I had just upgraded to gcc 3.3 from 3.1, and before the upgrade I never saw the problem. Is everyone who sees the problem using gcc 3.3, by any chance? If so then it's probably an incompatibility between curses.h and that compiler. |
|
|
| msg18279 - (view) |
Author: Anthony Baxter (anthonybaxter) |
Date: 2003-10-01 16:28 |
| Logged In: YES user_id=29957 Benno tells me that there's a patch for configure available at http://jeamland.net/~benno/configure.in.diff It'll fix FreeBSD, but not Mac OS X. Argh. |
|
|
| msg18280 - (view) |
Author: Anthony Baxter (anthonybaxter) |
Date: 2003-10-01 16:42 |
| Logged In: YES user_id=29957 Ok. Ripping out the line AC_DEFINE(_XOPEN_SOURCE_EXTENDED, 1, Define to activate Unix95-and-earlier features) from configure.in fixes the problem on FreeBSD. The only occurance of _XOPEN_SOURCE_EXTENDED in FreeBSD's headers is ncurses.h. FreeBSD will hopefully fix the bug in their headers to remove the problem in a future version of the code. Jack or Richard, can you see if _XOPEN_SOURCE_EXTENDED is used anywhere else on Mac OS X? If not, we could put a workaround in for Mac OS X and FreeBSD to not define the header. |
|
|
| msg18281 - (view) |
Author: Benno Rice (benno) |
Date: 2003-10-01 16:59 |
| Logged In: YES user_id=9925 There is a bug in FreeBSD's ncurses.h where it checks to see if a macro called __wchar_t is defined before attempting to typedef it's own wchar_t. This causes problems because __wchar_t (and wchar_t) exist, but __wchar_t is a typedef not a macro. This is solved by removing _XOPEN_SOURCE_EXTENDED which is what triggers this check. Removing _XOPEN_SOURCE_EXTENDED should have no other implications on FreeBSD. There is a patch that disables _XOPEN_SOURCE_EXTENDED for FreeBSD at http://jeamland.net/~benno/configure.in.diff |
|
|
| msg18282 - (view) |
Author: Jack Jansen (jackjansen) * |
Date: 2003-10-02 20:48 |
| Logged In: YES user_id=45365 On Jaguar _XOPEN_SOURCE_EXTENDED is only used in curses.h and ncurses.h, and one other place defines it (httpd/ap_config.h). The other good news is that on Panther the problem doesn't exist. |
|
|
| msg18283 - (view) |
Author: Anthony Baxter (anthonybaxter) |
Date: 2003-10-04 07:56 |
| Logged In: YES user_id=29957 Fixed in configure.in 1.438, 1.427.4.8; configure 1.428, 1.416.4.9 there's also a patch available from www.python.org/2.3.2/bugs.html for people who need it working on 2.3.2. |
|
|