Commit | Line | Data |
760ac839 |
1 | # This file has been put together by Anno Siegel <siegel@zrz.TU-Berlin.DE>, |
2 | # Andreas Koenig <k@franz.ww.TU-Berlin.DE> and Gerd Knops <gerti@BITart.com>. |
3 | # Comments, questions, and improvements welcome! |
4aa0a1f7 |
4 | # |
5 | # These hints work for NeXT 3.2 and 3.3. 3.0 has it's own |
6 | # special hint file. |
760ac839 |
7 | # |
4aa0a1f7 |
8 | |
da97b635 |
9 | ###################################################################### |
10 | # THE MALLOC STORY |
11 | ###################################################################### |
12 | # 1994: |
13 | # the simple program `for ($i=1;$i<38771;$i++){$t{$i}=123}' fails |
14 | # with Larry's malloc on NS 3.2 due to broken sbrk() |
15 | # |
16 | # setting usemymalloc='n' was the solution back then. Later came |
17 | # reports that perl would run unstable on 3.2: |
18 | # |
19 | # 1996: |
20 | # From about perl5.002beta1h perl became unstable on the |
21 | # NeXT. Intermittent coredumps were frequent on 3.2 OS. There were |
22 | # reports, that the developer version of 3.3 didn't have problems, so it |
23 | # seemed pretty obvious that we had to work around an malloc bug in 3.2. |
24 | # This hints file reflects a patch to perl5.002_01 that introduces a |
25 | # home made sbrk routine (remember, NeXT's sbrk _never_ worked). This |
26 | # sbrk makes it possible to run perl with its own malloc. Thanks to |
27 | # Ilya who showed me the way to his sbrk for OS/2!! |
28 | # |
29 | # The whole malloc desaster lead to a failing gdbm test. It is far |
30 | # beyond my understanding, why GDBM_File breaks with the "fix", but in |
31 | # general I consider it better to have a working perl with broken GDBM |
32 | # than no perl at all. |
33 | # |
34 | # So, this hintsfile is using perl's malloc. If you want to turn |
35 | # perl's malloc off, you need to remove '-DUSE_PERL_SBRK' and |
36 | # '-DHIDEMYMALLOC' from the ccflags and set usemymalloc to 'n'. |
37 | # |
38 | # 1997: |
39 | # From perl5.003_22 the malloc bug has no impact any more. We can run |
40 | # a perl without a special sbrk. Apparently Chip Salzenberg, the hero |
41 | # of 5.004 anyway, earned another trophy during Australien Open. |
42 | # |
43 | # use the following two lines to enable USE_PERL_SBRK. Try this if you |
44 | # encounter intermittent core dumps: |
45 | #ccflags='-DUSE_NEXT_CTYPE -DUSE_PERL_SBRK -DHIDEMYMALLOC' |
46 | #usemymalloc='y' |
47 | # use the following two lines if you have perl5.003_22 or better and |
48 | # do not encounter intermittent core dumps. |
49 | |
50 | ccflags='-DUSE_NEXT_CTYPE' |
51 | usemymalloc='n' |
52 | |
53 | ###################################################################### |
54 | # End of the MALLOC story |
55 | ###################################################################### |
56 | |
85e6fe83 |
57 | ldflags='-u libsys_s' |
a0d0e21e |
58 | libswanted='dbm gdbm db' |
5d94fbed |
59 | |
61d11634 |
60 | lddlflags='-nostdlib -r' |
5d94fbed |
61 | # Give cccdlflags an empty value since Configure will detect we are |
62 | # using GNU cc and try to specify -fpic for cccdlflags. |
63 | cccdlflags=' ' |
64 | |
760ac839 |
65 | # |
66 | # Change the line below if you do not want to build 'quad-fat' |
67 | # binaries |
68 | # |
40ea3355 |
69 | archs=`/bin/lipo -info /usr/lib/libm.a | sed 's/^[^:]*:[^:]*: //'` |
70 | for d in $archs |
71 | do |
72 | mab="$mab -arch $d" |
73 | done |
74 | |
61d11634 |
75 | archname='next-fat' |
76 | ld='cc' |
77 | |
a0d0e21e |
78 | i_utime='undef' |
79 | groupstype='int' |
80 | direntrytype='struct direct' |
81 | d_strcoll='undef' |
760ac839 |
82 | |
a0d0e21e |
83 | d_uname='define' |
2736da80 |
84 | # setpgid() is in the posix library, but we don't use -posix, so |
85 | # we don't see it. ext/POSIX/POSIX.xs *does* use -posix, so |
86 | # setpgid is still available as POSIX::setpgid. |
87 | # See ext/POSIX/POSIX/hints/next.pl. |
88 | d_setpgid='undef' |
a0d0e21e |
89 | d_setsid='define' |
90 | d_tcgetpgrp='define' |
91 | d_tcsetpgrp='define' |
760ac839 |
92 | |
ecfc5424 |
93 | # |
94 | # On some NeXT machines, the timestamp put by ranlib is not correct, and |
95 | # this may cause useless recompiles. Fix that by adding a sleep before |
96 | # running ranlib. The '5' is an empirical number that's "long enough." |
760ac839 |
97 | # |
ecfc5424 |
98 | ranlib='sleep 5; /bin/ranlib' |
760ac839 |
99 | |
40000a8c |
100 | # |
101 | # There where reports that the compiler on HPPA machines |
102 | # fails with the -O flag on pp.c. |
2736da80 |
103 | # Compiling pp.c with -O for HPPA machines results in a broken perl. |
104 | # This is true whether we're on an HPPA machine or cross-compiling |
105 | # for one. |
106 | pp_cflags='optimize=""' |