Commit | Line | Data |
---|---|---|
61ae2fbf JH |
1 | ########################################################################## |
2 | # *** README.mint | |
3 | ########################################################################## | |
4 | ||
5 | If you want to build perl yourself on MiNT (or maybe on an Atari without | |
6 | MiNT) you may want to accept some advice from somebody who already did it... | |
7 | ||
8 | There was a perl port for Atari ST done by ++jrb bammi@cadence.com. | |
9 | This port tried very hard to build on non-MiNT-systems. For the | |
10 | sake of efficiency I've left this way. Yet, I haven't removed bammi's | |
11 | patches but left them intact. Unfortunately some of the files that | |
12 | bammi contributed to the perl distribution seem to have vanished? | |
13 | ||
14 | So, how can you distinguish my patches from bammi's patches? All of | |
15 | bammi's stuff is embedded in "#ifdef atarist" preprocessor macros. | |
16 | My MiNT port uses "#ifdef __MINT__" instead (and unconditionally | |
17 | undefines "atarist". If you want to continue on bammi's port, all | |
18 | you have to do is to swap the "-D" and "-U" switches for "__MINT__" | |
19 | and "atarist" in the variable ccflags. | |
20 | ||
21 | However, I think that my version will still run on non-MiNT-systems | |
22 | provided that the user has a Eunuchs-like environment (i.e. the | |
23 | standard envariables like $PATH, $HOME, ... are set, there is a | |
24 | POSIX compliant shell in /bin/sh, and...) | |
25 | ||
26 | Known problems | |
27 | ============== | |
28 | ||
29 | The problems you may encounter when building perl on your machine | |
30 | are most probably due to deficiencies in MiNT resp. the Atari | |
31 | platform in general. | |
32 | ||
33 | First of all, if you have less than 8 MB of RAM you shouldn't | |
34 | even try to build Perl yourself. Better grab a binary pre-compiled | |
35 | version somewhere. Even if you have more memory you should take | |
36 | some care. Try to run in a fresh environment (without memory | |
37 | fragmented too much) with as few daemons, accessories, xcontrol | |
38 | modules etc. as possible. If you run some AES you should | |
39 | consider to start a console based environment instead. | |
40 | ||
41 | A problem has been reported with sed. Sed is used to create | |
42 | some configuration files based on the answers you have given | |
43 | to the Configure script. Unfortunately the Perl Configure script | |
44 | shows sed on MiNT its limits. I have sed 2.05 with a stacksize | |
45 | of 64k and I have encountered no problems. If sed crashes | |
46 | during your configuration process you should first try to | |
47 | augment sed's stacksize: | |
48 | ||
49 | fixstk 64k /usr/bin/sed | |
50 | ||
51 | (or similar). If it still doesn't help you may have a look | |
52 | which other versions of sed are installed on your system. | |
53 | If you have a KGMD 1.0 installation you will find three | |
54 | in /usr/bin. Have a look there. | |
55 | ||
56 | Perl has some "mammut" C files. If gcc reports "internal | |
57 | compiler error: program cc1 got fatal signal 10" this is very | |
58 | likely due to a stack overflow in program cc1. Find cc1 | |
59 | and fix its stack. I have made good experiences with | |
60 | ||
61 | fixstk 2 cc1 | |
62 | ||
63 | This doesn't establish a stack of 2 Bytes only as you might | |
64 | think. It really reserves one half of the available memory | |
65 | for cc1's stack. A setting of 1 would reserve the entire | |
66 | memory for cc1, 3 would reserve three thirds. You will have | |
67 | to find out the value that suits to your system yourself. | |
68 | ||
69 | BTW, cc1 is maybe a little hard to find. It is generally installed | |
70 | as | |
71 | /usr/local/lib/gcc-lib/<platform>/<gcc-version>/cc1 | |
72 | ||
73 | where <platform> is probably something like "m68k-atari-mint" | |
74 | and <version> is the gcc version you use (find out with | |
75 | "gcc --version"). Maybe "gcc-lib" is not installed in | |
76 | "/usr/local/lib" but "/usr/lib" on your system. | |
77 | ||
78 | Now run make (maybe "make -k"). If you get a fatal signal 10 | |
79 | increase cc1's stacksize, if you run out of memory you should | |
80 | either decrease the stacksize or follow some more hints: | |
81 | ||
82 | Perl's building process is very handy on machines with a lot | |
83 | of virtual memory but may result in a desaster if you are short | |
84 | of memory. If gcc fails to compile many source files you should | |
85 | reduce the optimization. Grep for "optimize" in the file | |
86 | config.sh and change the flags. | |
87 | ||
88 | If only several huge files cause problems (actually it is not a | |
89 | matter of the file size resp. the amount of code but depends on | |
90 | the size of the individual funtions) it is useful to bypass | |
91 | the make program and compile these files directly from the | |
92 | command line. For example if you got something like the | |
93 | following from make: | |
94 | ||
95 | CCCMD = gcc -DPERL_CORE .... | |
96 | ... | |
97 | ...: virtual memory exhausted | |
98 | ||
99 | you should hack into the shell: | |
100 | ||
101 | gcc -DPERL_CORE ... toke.c | |
102 | ||
103 | Please note that you have to add the name of the source file | |
104 | (here toke.c) at the end. | |
105 | ||
106 | If none of this helps, you're helpless. Wait for a binary | |
107 | release. If you have succeded you may encounter another problem | |
108 | at the linking process. If gcc complains that it can't find | |
109 | some libraries within the perl distribution you probably have | |
110 | an old linker. If it complains for example about "file not | |
111 | found for xxx.olb" you should cd into the directory in | |
112 | question and | |
113 | ||
114 | ln -s libxxx.a xxx.olb | |
115 | ||
116 | This will fix the problem. | |
117 | ||
118 | This version (5.00402) of perl has passed most of the tests on my system: | |
119 | ||
120 | Failed Test Status Wstat Total Fail Failed List of failed | |
121 | ------------------------------------------------------------------------------ | |
122 | io/pipe.t 10 2 20.00% 7, 9 | |
123 | io/tell.t 13 1 7.69% 12 | |
124 | lib/complex.t 762 13 1.71% 84-85, 248-251, 257, 272-273, | |
125 | 371, 380, 419-420 | |
126 | lib/io_pipe.t 10 1 10.00% 9 | |
127 | lib/io_tell.t 13 1 7.69% 12 | |
128 | op/magic.t 30 2 6.67% 29-30 | |
129 | Failed 6/152 test scripts, 96.05% okay. 20/4359 subtests failed, 99.54% okay. | |
130 | ||
131 | Pipes always cause problems with MiNT, it's actually a surprise that | |
132 | most of the tests did work. I've got no idea why the "tell" test failed, | |
133 | this shouldn't mean too big a problem however. | |
134 | ||
135 | Most of the failures of lib/complex seem to be harmless, actually errors | |
136 | far right to the decimal point... Two failures seem to be serious: | |
137 | The sign of the results is reversed. I would say that this is due | |
138 | to minor bugs in the portable math lib that I compiled perl with. | |
139 | ||
140 | I haven't bothered very much to find the reason for the failures | |
141 | with op/magic.t and op/stat.t. Maybe you'll find it out. | |
142 | ||
143 | ########################################################################## | |
144 | ||
145 | Another possible problem may arise from the implementation of the "pwd" | |
146 | command. It happened to add a carriage return and newline to its output | |
147 | no matter what the setting of $UNIXMODE is. This is quite annoying since many | |
148 | library modules for perl take the output of pwd, chop off the | |
149 | trailing newline character and then expect to see a valid path in | |
150 | that. But the carriage return (last but second character!) isn't | |
151 | chopped off. You can either try to patch all library modules (at | |
152 | the price of performance for the extra transformation) or you can | |
153 | use my version of pwd that doesn't suffer from this deficiency. | |
154 | ||
155 | The fixed implementation is in the mint subdirectory. Running | |
156 | "Configure" will attempt to build and install it if necessary | |
157 | (hints/mint.sh will do this work) but you can build and install it | |
158 | explicitly by: | |
159 | ||
160 | cd mint | |
161 | make install | |
162 | ||
163 | This is the fastest solution. | |
164 | ||
165 | Just in case you want to go the hard way: perl won't even build with a | |
166 | broken pwd! You will have to fix the library modules | |
167 | (ext/POSIX/POSIX.pm, lib/Cwd.pm, lib/pwd.pl) at last after building | |
168 | miniperl. | |
169 | ||
170 | A major nuisance of current MiNTLib versions is the implementation | |
171 | of system() which is far from being POSIX compliant. A real system() | |
172 | should fork and then exec /bin/sh with its argument as a command | |
173 | line to the shell. The MiNTLib system() however doesn't expect | |
174 | that every user has a POSIX shell in /bin/sh. It tries to work | |
175 | around the problem by forking and exec'ing the first token in its argument | |
176 | string. To get a little bit of compliance to POSIX system() it | |
177 | tries to handle at least redirection ("<" or ">") on its own | |
178 | behalf. | |
179 | ||
180 | This isn't a good idea since many programs expect that they can | |
181 | pass a command line to system() that exploits all features of a | |
182 | POSIX shell. If you use the MiNTLib version of system() with | |
183 | perl the Perl function system() will suffer from the same deficiencies. | |
184 | ||
185 | You will find a fixed version of system() in the mint subdirectory. | |
186 | You can easily insert this version into your system libc: | |
187 | ||
188 | cd mint | |
189 | make system.o | |
190 | ar r /usr/lib/libc.a | |
191 | ranlib /usr/lib/libc.a | |
192 | ||
193 | If you are suspicious you should either back up your libc before | |
194 | or extract the original system.o from your libc with | |
195 | "ar x /usr/lib/libc.a system.o". You can then backup the system.o | |
196 | module somewhere before you succeed. | |
197 | ||
198 | Anything missing? Yep, I've almost forgotten... | |
199 | No file in this distribution without a fine saying. Take this one: | |
200 | ||
201 | "From a thief you should learn: (1) to work at night; | |
202 | (2) if one cannot gain what one wants in one night to | |
203 | try again the next night; (3) to love one's coworkers | |
204 | just as thieves love each other; (4) to be willing to | |
205 | risk one's life even for a little thing; (5) not to | |
206 | attach too much value to things even though one has | |
207 | risked one's life for them - just as a thief will resell | |
208 | a stolen article for a fraction of its real value; | |
209 | (6) to withstand all kinds of beatings and tortures | |
210 | but to remain what you are; and (7) to believe your | |
211 | work is worthwhile and not be willing to change it." | |
212 | ||
213 | -- Rabbi Dov Baer, Maggid of Mezeritch | |
214 | ||
215 | OK, this was my motto while working on Perl for MiNT, especially rule (1)... | |
216 | ||
217 | Have fun with Perl! | |
218 | ||
219 | Guido Flohr | |
220 | -- | |
221 | mailto:gufl0000@stud.uni-sb.de | |
222 | http://stud.uni-sb.de/~gufl0000 |