Re: Fixing broken djgpp support in make 3.81
Hello Paul: [snip] It seems like the amount of change you've made will require copyright assignment. I see you've already done some of these, for Bison etc., but if you need paperwork let me know. Please send the required paperwork to me or let me known where I can get it and I will sign the copyright assignment. As for the above, I wonder whether it might be useful to add: # define HAVE_DRIVE(n)0 for the non-HAVE_DOS_PATHS sections of the above ifdef. That would allow us to use HAVE_DRIVE(foo) without enclosing it in ifdefs for DOS etc., and still have the compiler omit the code (via optimization of always-false values). You are right and that is the way i would do it but an inspection of the make code showed me that all code lines where the macro can be used were enclosed by ifdefs for DOS, so I have assumed that this was the prefered coding style and in such a case it made no much sense to me to define macros that would never be used for other OSs. If some day someone decides to criticaly review the existing code, one of the goals should be to get rid of all these #ifdef HAVE_DOS_PATHS and VMS (if possible at all) and to replace that code with some generic path syntax handling code that will make use of such macros to hide OS path syntax peculiarities. Then it will become responsability of the different port maintainers to provide adecuate macros to handle the issue. I resend the patch with included empty/false definitions of HAVE_DRIVE for the other OSs in make.h and also fixed character set issues I have noticed to late in the changelog text. Regards, Juan M. Guerrero 2008-01-15 Juan Manuel Guerrero [EMAIL PROTECTED] * implicit.c: Add const qualifier to *lastslash so it matches the declaration of *filename. * dir.c (file_exists_p) [HAVE_DOS_PATHS]: Use HAVE_DRIVE. (file_impossible) [HAVE_DOS_PATHS]: Use HAVE_DRIVE. (file_impossible_p) [HAVE_DOS_PATHS]: Use HAVE_DRIVE. * function.c [HAVE_DOS_PATHS, VMS]: Remove IS_PATHSEP definition. Now it is defined in its canonical place in make.h. (abspath) [HAVE_DOS_PATHS]: Use IS_PATHSEP instead of a literal '/'. Support d:foo style absolute file names. Add logic for the case that backslash are path separators too. Include colon as separator char in the IS_PATHSEP definition and use it to check for root char. (func_notdir_suffix, func_basename_dir, abspath): Use HAVE_DRIVE. * glob/glob.c (my_realloc): Don't define, and don't redefine realloc to call it since the DJGPP realloc handles NULL pointers. * hash.c (round_up_2): Use 4294967295U to avoid compiler warnings. * implicit.c (pattern_search) [HAVE_DOS_PATHS]: Use HAVE_DRIVE. * make.h: All OS specific macros to handle path syntax peculiarities are defined here. [HAVE_DOS_PATHS]: New macro HAVE_DRIVE to check for drive specifier strings like 'a:'. For all other supported OSs this macro is a no operation. [HAVE_DOS_PATHS, VMS]: Define macro IS_PATHSEP. Previous definition in function.c removed accordingly. * Makefile.DOS.template (INCLUDES): Use $(prefix) and the corresponding variables to define LIBDIR, INCLUDEDIR and LOCALEDIR instead of using the hardcoded ones. (SUBDIRS): doc subdir added. (INFO_DEPS, DVIS): Values changed to 'make.info' and 'make.dvi'. (TEXI2HTML, TEXI2HTML_FLAGS): Removed. Use makeinfo --html to create html formated docs. texi2html may not be ported to DOS. (make.info, make.dvi, make.ps, make.html): Make targets depend on 'make.texi'. (.texi.info, .texi, .texi.dvi): Now recursively invocate. Change -I switch to look in ./ instead of ./doc. (html): Target depend on html-recursive instead of make_1.html. (make_1.html): Removed. (mostlyclean-aminfo): Use $(srcdir)/doc instead of ./ as prefix. (all-recursive): Allow for more than one subdir in the build process. (mostlyclean-recursive, clean-recursive, distclean-recursive, maintainer-clean-recursive, check-recursive): Enter in doc/ too. (tags-recursive): Allow for more than one subdir in the build process. (info-recursive, dvi-recursive, ps-recursive, html-recursive): New targets. Enter into doc/ to produce the targets. (all-am): $(INFO_DEPS) replaced by info. * dosbuild.bat: Add -DLOCALEDIR to compiler command line to compile main.c. Remove make.new if build successfull. Use /dev/env/DJDIR instead of c:/djgpp for -DINCLUDEDIR and -DLOCALEDIR. * tests/scripts/functions/abspath: Added new MSDOS specific tests. They reproduce the complete test cases adding up to three new test cases but using backslash instead of slash as dir separator and prepending a DOS typical
Re: Fixing broken djgpp support in make 3.81
On Tue, 2008-01-15 at 01:32 +0100, Juan Manuel Guerrero wrote: /* Handle other OSs. */ #ifndef PATH_SEPARATOR_CHAR # if defined(HAVE_DOS_PATHS) # define PATH_SEPARATOR_CHAR ';' # define IS_PATHSEP(c) ((c) == '/' || (c) == '\\') # define HAVE_DRIVE(n) ((n)[0] (n)[1] == ':') # elif defined(VMS) # define PATH_SEPARATOR_CHAR ',' # define IS_PATHSEP(c) ((c) == ']') # else # define PATH_SEPARATOR_CHAR ':' # define IS_PATHSEP(c) ((c) == '/') # endif #endif Hi Juan; it's great to have someone paying some attention to this port again. It seems like the amount of change you've made will require copyright assignment. I see you've already done some of these, for Bison etc., but if you need paperwork let me know. As for the above, I wonder whether it might be useful to add: # define HAVE_DRIVE(n)0 for the non-HAVE_DOS_PATHS sections of the above ifdef. That would allow us to use HAVE_DRIVE(foo) without enclosing it in ifdefs for DOS etc., and still have the compiler omit the code (via optimization of always-false values). -- --- Paul D. Smith [EMAIL PROTECTED] Find some GNU make tips at: http://www.gnu.org http://make.mad-scientist.us Please remain calm...I may be mad, but I am a professional. --Mad Scientist ___ Bug-make mailing list Bug-make@gnu.org http://lists.gnu.org/mailman/listinfo/bug-make
Re: Fixing broken djgpp support in make 3.81
From: Juan Manuel Guerrero [EMAIL PROTECTED] Date: Thu, 15 Nov 2007 05:39:46 +0100 Cc: I have appended a small patch that shows the things I had to change to get a working make binary. That patch is not intended to be used. It shall only show the code lines that may need to be adjusted to solve the problems. There may be better ways to solve the issue, but I am not familiar enough with the make sources to provide a better patch. There may be also much more MSDOS or DJGPp specific things that may be fixed. The patch only fixes things that I am aware. Things that may cause trouble are drive letters, checks for absolute patch, etc, etc etc. The usual DOS/WINDOWS things. Thanks, Juan. Your warnings notwithstanding, I did install your patches. I no longer have time to do any significant work on the DJGPP port of Make, so installing your changes will at least give DJGPP users a binary that works better than with the original sources. I have not waiste my time trying to fix the Makefile.DOS, but now that make.texi has been moved into /doc it needs to be adjusted accordingly. It would be nice if rules for dvi, ps, pdf and html could be added. It would also be usefull not to split the MSDOS version of make.info anymore. The extensions .info-[0-9] are changed to .i[0-9] to acomplish with the 8.3 file name rule but it is always forgotten to adjust the file name strings in make.info accordingly. All this difficulties can be avoided by no spliting the info formated docs anymore. Please note that dvi and ps are already supported, and pdf is not supported on any platform. I made all the other changes you suggested; please see the patches below. Thanks again for your efforts supporting DJGPP ports in general, and Make in particular. 2007-12-22 Eli Zaretskii [EMAIL PROTECTED] Suggested by Juan Manuel Guerrero [EMAIL PROTECTED]: * Makefile.DOS.template (info_TEXINFOS): Remove unused variable. (TEXINFOS): Value changed to `doc/make.texi'. (.SUFFIXES): Use .texi instead of .texinfo. (make.info, make.dvi): Depend on doc/make.texi. (.texi.info): New target, instead of .texinfo.info. Change -I switch to $(MAKEINFO) to look in doc/. Use --no-split. (.texi): New target, instead of .texinfo. Change -I switch to $(MAKEINFO) to look in doc/. Use --no-split. (.texi.dvi): New target, instead of .texinfo.dvi. Change -I switch to $(MAKEINFO) to look in doc/. (install-info-am, uninstall-info): Don't look for *.i[0-9] and *.i[0-9][0-9] (due to --no-split above). (noinst_TEXINFOS, TEXI2HTML, TEXI2HTML_FLAGS): New variables. (html, make_1.html): New targets. (.PHONY): Add html. (.SUFFIXES): Add .html. Index: Makefile.DOS.template === RCS file: /sources/make/make/Makefile.DOS.template,v retrieving revision 2.15 diff -u -r2.15 Makefile.DOS.template --- Makefile.DOS.template 4 Jul 2007 19:35:16 - 2.15 +++ Makefile.DOS.template 22 Dec 2007 12:04:40 - @@ -83,7 +83,6 @@ libglob_a_SOURCES =%GLOB_SOURCES% make_LDADD = glob/libglob.a -info_TEXINFOS =make.texinfo man_MANS = make.1 INCLUDES = -I$(srcdir)/glob -DLIBDIR=\c:/djgpp/lib\ -DINCLUDEDIR=\c:/djgpp/include\ -DLOCALEDIR=\$(localedir)\ @@ -118,10 +117,14 @@ TEXINFO_TEX = $(srcdir)/config/texinfo.tex INFO_DEPS = make.info DVIS = make.dvi -TEXINFOS = make.texinfo +TEXINFOS = doc/make.texi +noinst_TEXINFOS = doc/fdl.texi doc/make-stds.texi man1dir = $(mandir)/man1 MANS = $(man_MANS) +TEXI2HTML = texi2html +TEXI2HTML_FLAGS = -split_chapter + NROFF = nroff DIST_COMMON = README ABOUT-NLS AUTHORS COPYING ChangeLog INSTALL Makefile.am Makefile.in NEWS acconfig.h aclocal.m4 alloca.c build.sh-in config.h-in configure configure.in getloadavg.c @@ -136,7 +139,7 @@ default: all .SUFFIXES: -.SUFFIXES: .c .dvi .info .o .obj .ps .texinfo .tex +.SUFFIXES: .c .dvi .info .o .obj .ps .texi .tex .html mostlyclean-hdr: @@ -186,33 +189,41 @@ @command.com /c if exist make.exe del make.exe $(LINK) $(make_LDFLAGS) $(make_OBJECTS) $(make_LDADD) $(LIBS) -make.info: make.texinfo -make.dvi: make.texinfo +# Documentation + +make.info: doc/make.texi +make.dvi: doc/make.texi DVIPS = dvips -.texinfo.info: +.texi.info: @command.com /c if exist make.info* del make.info* @command.com /c if exist make.i* del make.i* - $(MAKEINFO) -I$(srcdir) $ -o ./$@ + $(MAKEINFO) -I$(srcdir)/doc --no-split $ -o ./$@ -.texinfo: +.texi: @command.com /c if exist make.info* del make.info* @command.com /c if exist make.i* del make.i* - $(MAKEINFO) -I$(srcdir) $ -o ./$@ - -.texinfo.dvi: - TEXINPUTS=$(srcdir);$$TEXINPUTSMAKEINFO='$(MAKEINFO) -I $(srcdir)' $(TEXI2DVI) $ + $(MAKEINFO) -I$(srcdir)/doc --no-split $ -o ./$@
Fixing broken djgpp support in make 3.81
The djgpp specific support of make 3.81 is brocken. There are two sources for this: 1) failure of the compilation stage. It is not possible to compile the sources out-of-the-box because: - configh.dos defines HAVE_SYS_SIGLIST instead of HAVE_DECL_SYS_SIGLIST what is the macro name that is checked for in signame.c - in job.c the function child_execute_job is compiled when __MSDOS__ is defined. This is wrong; the DJGPP port uses directly system() and spawnvpe() in start_job_command. There is no MSDOS/DJGPP version of child_execute_job at all. These two issues had to been fixed before it was possible to compile the successfully the sources. There is also a mismatch of the function prototype of my_realloc in glob.c with realloc() of djgpp's libc, but this only causes a compiler warning and does not stop the show. Neitherless it is woth to be fixed; see patch. 2) failure of the produced binary when used with an OS that does not use bash as default shell. After fixing configh.dos and inhibiting the use of child_execute_job() in job.c I was able to produce a binary but this binary did not work as expected. I tried to run the testsuite by typing the command: make check and I got the following output on the screen: Befehl oder Dateiname nicht gefunden (command or file not found. Error message from command.com started by system() of libc) Befehl oder Dateiname nicht gefunden Befehl oder Dateiname nicht gefunden Befehl oder Dateiname nicht gefunden No command name seen. (Error message from djgpp's libc system() call) make.exe: *** [check-recursive] Error -1 The text in parenthesis have been added by me. The string that is passed to djgpp's libc system() that originate the above error messages is: failcom='exit 1'; \ for f in x $MAKEFLAGS; do \ case $f in \ *=* | --[!k]*);; \ *k*) failcom='fail=yes';; \ esac; \ done; \ dot_seen=no; \ target=`echo check-recursive | sed s/-recursive//`; \ list='glob config po doc '; for subdir in $list; do \ echo Making $target in $subdir; \ if test $subdir = .; then \ dot_seen=yes; \ local_target=$target-am; \ else \ local_target=$target; \ fi; \ (cd $subdir e:/p2/make-3.81/make.exe $local_target) \ || eval $failcom; \ done; \ if test $dot_seen = no; then \ e:/p2/make-3.81/make.exe $target-am || exit 1; \ fi; test -z $fail The average user will _never_ figure out why the same makefile fails so miserably when make 3.81 is used and works perfectly when make 3.80 is used. It took me some hours to understand what had happend. For some reason completly unclear to me, some one has decided not to propagate the value of the SHELL environment variable to the child process anymore. In the case of MSDOS/DJGPP this means that the value of SHELL is deleted from the environment before system() is called. On _all_ non-POSIX operating systems this will have catastrophic implications because the libc system() function will have to use the default shell or command processor. The default shell on non-POSIX OSs is usualy not bash so the above shell script must fail. After some stepping through the code with gdb I have found that the problem can be fixed in define_automatic_variables() in variable.c. If compiled with DJGPP then SHELL will always be exported. Please note that I am not objecting the change that have been done by the make maintainers. There may be some good reason not to export automatically all the envorenment variables. Neitherless if make shall continue supporting non-PoSIX platforms, certain precaution should be taken. In this particular case it has been shown that it is not valid to assume that the used OS will provide a POSIX compatible shell or command processor. If no one takes care of things like this issue, make becomes useless on certain platforms. I have appended a small patch that shows the things I had to change to get a working make binary. That patch is not intended to be used. It shall only show the code lines that may need to be adjusted to solve the problems. There may be better ways to solve the issue, but I am not familiar enough with the make sources to provide a better patch. There may be also much more MSDOS or DJGPp specific things that may be fixed. The patch only fixes things that I am aware. Things that may cause trouble are drive letters, checks for absolute patch, etc, etc etc. The usual DOS/WINDOWS things. I have also appended the output generated by the testsuite. It shows that there are some more things to fix. I have not waiste my time trying to fix the Makefile.DOS, but now that make.texi has been moved into /doc it needs to be adjusted accordingly. It would be nice if rules for dvi, ps, pdf and html could be added. It would also be usefull not to split the MSDOS version of make.info anymore. The extensions .info-[0-9] are changed to .i[0-9] to acomplish with the
