You mean patch-551 (2.95g) fails and patch-552 passes?

Well, no. patch-551 works, the tarball extracted from 2.95g does not.

Diffing an export of smalltalk@sv.gnu.org/smalltalk--devo--2.2--patch-551 (from savannah) against gst 2.95g tarball, I get some differences. ChangeLog entries, examples, snprintf tests and Unicode methods missing from the tarball. On the other side: match.h, prims.inl, vm.inl are not in the tla repo, so I suppose they get regenerated at compile time. Maybe that's the culprit? Or maybe that's an autotools issue?

Removing *.inl and match.h does not help.

I'll look at the diffs, can you please check if there are differences in config.log or similar between patch-551 and 2.95g?

autoreconf'ing does not help either.

The diff is available here: http://thomas.g.girard.free.fr/gst/config.log_tarball_to_patch-551.diff

(Because of the autoreconf, lines numbers are different. I'll do it again tomorrow.)

Full config.log files available from the same directory.

I've updated the diff, there's nothing interesting in it.

Here's the backtrace in gdb I get after a handle SIGSEGV nostop noprint pass no to interfere with libsigsegv:

#0  0xb7d0aea6 in raise () from /lib/libc.so.6
No symbol table info available.
#1  0xb7d0c7b1 in abort () from /lib/libc.so.6
No symbol table info available.
#2  0xb7ee52df in oldspace_sigsegv_handler (fault_address=0x4, serious=0)
    at oop.c:942
        page = (void *) 0x0
        reentering = 0
        reentered = 0
#3  0xb7e2dcf4 in sigsegv_handler (sig=11, more=51) at handler-unix.c:134
        address = (void *) 0x4
#4  <signal handler called>
No symbol table info available.
#5  0xb7f222de in _gst_interpret (processOOP=0x40317760) at vm.def:640
        _receiver = (OOP) 0x40352b58
        _stack0 = (OOP) 0xb7b00748
        _stack1 = (OOP) 0x40352b58
        ip = (ip_type) 0xb7b0074c "3"
        sp = (OOP *) 0x8064118
        arg = 3
        monitored_byte_codes = {0xb7f1bca2 <repeats 256 times>}
        true_byte_codes = {0xb7f1be63 <repeats 42 times>, 0xb7f1d8ee, 
  0xb7f1dc86, 0xb7f1be63, 0xb7f1be63, 0xb7f1be63, 0xb7f1be63, 0xb7f1be63, 
  0xb7f1be63, 0xb7f1be63, 0xb7f1be63, 0xb7f1be2f, 
  0xb7f1be63 <repeats 203 times>}
        false_byte_codes = {0xb7f1be11 <repeats 42 times>, 0xb7f1dc86, 
  0xb7f1d8ee, 0xb7f1be11, 0xb7f1be11, 0xb7f1be11, 0xb7f1be11, 0xb7f1be11, 
  0xb7f1be11, 0xb7f1be11, 0xb7f1be11, 0xb7f1bddd, 
  0xb7f1be11 <repeats 203 times>}
        normal_byte_codes = {0xb7f1bd7b, 0xb7f1e980, 0xb7f1e688, 0xb7f1e62c, 
  0xb7f1e5d0, 0xb7f1ea45, 0xb7f1e766, 0xb7f1e9e9, 0xb7f1e4b3, 0xb7f1e370, 
  0xb7f1e298, 0xb7f1e313, 0xb7f1e207, 0xb7f1e18c, 0xb7f1e12f, 0xb7f1e0d2, 
  0xb7f1e010, 0xb7f1e8a2, 0xb7f1e850, 0xb7f1e7f8, 0xb7f1e7c2, 0xb7f1ed49, 
  0xb7f1e920, 0xb7f1eb16, 0xb7f1e596, 0xb7f1ebd3, 0xb7f1eb7d, 0xb7f1ec25, 
  0xb7f1e6e1, 0xb7f1ecd3, 0xb7f1eaa1, 0xb7f1ec6e, 0xb7f1d712, 0xb7f1dbe5, 
  0xb7f1db20, 0xb7f1dadd, 0xb7f1daaf, 0xb7f1da60, 0xb7f1d972, 0xb7f1d92d, 
  0xb7f1d90d, 0xb7f1d8ee, 0xb7f1d8b0, 0xb7f1d872, 0xb7f1d837, 0xb7f1d7f3, 
  0xb7f1d7b3, 0xb7f1d776, 0xb7f1d752, 0xb7f1ddcd, 0xb7f1dd84, 0xb7f1dd43, 
  0xb7f1dd18, 0xb7f1dca7, 0xb7f1dc86, 0xb7f1dc69, 0xb7f1dc2c, 0xb7f1deec, 
  0xb7f1dea3, 0xb7f1de5a, 0xb7f1de11, 0xb7f1df7e, 0xb7f1df35, 0xb7f1dfc7, 
  0xb7f1d6a9, 0xb7f1d276, 0xb7f1d231, 0xb7f1d1f4, 0xb7f1d1d0, 0xb7f1d394, 
  0xb7f1d2de, 0xb7f1d32b, 0xb7f1cf7b, 0xb7f1cef8, 0xb7f1ce86, 0xb7f1ce3e, 
  0xb7f1cdbd, 0xb7f1cd7e, 0xb7f1cd41, 0xb7f1ccf5, 0xb7f1cca0, 0xb7f1d562, 
  0xb7f1d49e, 0xb7f1d188, 0xb7f1d0f0, 0xb7f1d0b0, 0xb7f1d06c, 0xb7f1d00a, 
  0xb7f1cfa1, 0xb7f1d632, 0xb7f1d5c1, 0xb7f1d456, 0xb7f1d3d3, 0xb7f1cc46, 
  0xb7f1cbce, 0xb7f1cba5, 0xb7f1cb7f, 0xb7f1c897, 0xb7f1c813, 0xb7f1c7aa, 
  0xb7f1c749, 0xb7f1cb3d, 0xb7f1c90c, 0xb7f1cabf, 0xb7f1c68b, 0xb7f1c9b2, 
  0xb7f1c972, 0xb7f1ca47, 0xb7f1c6f2, 0xb7f1c930, 0xb7f1c608, 0xb7f1c5a5, 
  0xb7f1c563, 0xb7f1c515, 0xb7f1c4d3, 0xb7f1c3d3, 0xb7f1c34c, 0xb7f1c450, 
  0xb7f1c30f, 0xb7f1c294, 0xb7f1c25c, 0xb7f1c212, 0xb7f1c194, 0xb7f1c161, 
  0xb7f1c116, 0xb7f1c0ea, 0xb7f1c0bf, 0xb7f1c077, 0xb7f1c02f, 0xb7f203c4, 
  0xb7f20380, 0xb7f202f2, 0xb7f2022d, 0xb7f201d0, 0xb7f201b0, 0xb7f200de, 
  0xb7f200a4, 0xb7f205a7, 0xb7f20567, 0xb7f2061f, 0xb7f20482, 0xb7f20658, 
  0xb7f204fa, 0xb7f1fcad, 0xb7f1fc35, 0xb7f20c44, 0xb7f20c07, 0xb7f20c82, 
  0xb7f207d7, 0xb7f20d70, 0xb7f20b63, 0xb7f20cdd, 0xb7f1ff5b, 0xb7f20a64, 
  0xb7f209e1, 0xb7f20ae3, 0xb7f206b3, 0xb7f2096e, 0xb7f208fa, 0xb7f2099a, 
  0xb7f1f42f, 0xb7f1f9d4, 0xb7f1f98c, 0xb7f1f94a, 0xb7f1f8cf, 0xb7f1faa1, 
  0xb7f1fa75, 0xb7f1f892, 0xb7f1f80b, 0xb7f1f7a2, 0xb7f1f734, 0xb7f1f6bc, 
  0xb7f1f680, 0xb7f1f655, 0xb7f1f5c9, 0xb7f1f593, 0xb7f1f513, 0xb7f1fb70, 
  0xb7f1fae5, 0xb7f1f2c9, 0xb7f1f241, 0xb7f1f208, 0xb7f1f189, 0xb7f1f0ef, 
  0xb7f1f068, 0xb7f1f023, 0xb7f1efe1, 0xb7f1ef56, 0xb7f1eedc, 0xb7f1ee99, 
  0xb7f1ee1f, 0xb7f1edcb, 0xb7f1ed7f, 0xb7f21b5c, 0xb7f21990, 0xb7f21538, 
  0xb7f214b6, 0xb7f2188c, 0xb7f2180e, 0xb7f21af0...}
#6  0xb7f26592 in _gst_nvmsg_send (receiver=0x403121f0, 
    sendSelector=0x40319058, args=0xbfa698c4, sendArgs=1) at interp.c:2211
        processOOP = (OOP) 0x40317760
        currentProcessOOP = (OOP) 0x403542f0
        result = <value optimized out>
        i = <value optimized out>
#7  0xb7f01080 in _gst_va_msg_sendf (resultPtr=0x0, 
    fmt=0xb7f429da "%v %o changed: %S", ap=0xbfa69ad0 "") at callin.c:293
        selector = (OOP) 0x40319058
        args = (OOP *) 0xbfa698c0
        result = <value optimized out>
        i = 1
        numArgs = <value optimized out>
        fp = 0xb7f429eb ""
        s = <value optimized out>
        selectorBuf = "changed:", '\0' <repeats 68 times>, "ѽԷ", '\0' <repeats 12 times>, "ѽԷ", '\0' <repeats 16 times>, "���\000\000\000\000p��ѽԷ \000\000\000\004\000\000\0008�\005\b\000\000\000\000 \000\000\000\004\000\000\000��\005\b\000\000\000\000���\000\000\000\000��\000\000\000\000\000\000\210\000\000\000\021\000\000\000��\000\0000\000\000\000\000\000\000\000p��\000\000\000\0000\000\000\000\000\000\000\000�\217�\000\000\000\000@��x\232��-�Է@��\030\000\000\000\234<�\020�\005\b\030\000\000\000\000\0301@"
#8  0xb7f018e3 in _gst_msg_sendf (resultPtr=0x0, 
    fmt=0xb7f429da "%v %o changed: %S") at callin.c:368
        ap = 0xbfa69ac8 "�!1@�)�"
#9  0xb7ed824b in _gst_invoke_hook (hook=GST_RETURN_FROM_SNAPSHOT)
    at comp.c:539
        save_execution = 0
#10 0xb7ec911a in _gst_initialize (kernel_dir=0x0, 
    image_file=0xbfa6baf2 "/home/tgg/src/smalltalk-2.95g/gst.im", 
    flags=<value optimized out>) at files.c:516
        willRegressTest = 3215375120
        result = <value optimized out>
        currentDirectory = 0x804c008 "/home/tgg/src/smalltalk-2.95g"
        home = <value optimized out>
        str = 0x804e300 "/home/tgg/src/smalltalk-2.95g/gst.im"
        abortOnFailure = true
        rebuild_image_flags = 0
#11 0x08049102 in main (argc=54595653, argv=0x33) at main.c:372
        result = <value optimized out>
        file = <value optimized out>

Please let me know if you need something else.

> I've updated the diff, there's nothing interesting in it.

... so you're saying that the build is not deterministic and from the same
source code (except for some missing ChangeLogs :-) ) you got different
results.

Can you build 2.95g and patch-551 in two different directories, and see which
.o/.a/.la differ? This is quite disconcerting, if you weren't the reporter
I would have already closed it as invalid or worksforme...

It seems the gst.im gets corrupted after the first invocation.

How to reproduce it on my i386 box:

$ ./configure --with-tcl=/usr/lib/tcl8.4 --with-tk=/usr/lib/tk8.4 && make
....
$ ls -l gst.im
-rw-r--r-- 1 tgg tgg 1630820 Dec 14 16:06 gst.im
$ cp gst.im gst.im-
$ ./gst -S --image=./gst.im tests/AnsiLoad.st
"Global garbage collection... done"
Loading package SUnit
Loading Ansi.st
Loading AnsiDB.st
Loading AnsiInit.st
"Global garbage collection... done"
$ ls -l gst.im*
-rw-r--r-- 1 tgg tgg 2827244 Dec 14 16:14 gst.im
-rw-r--r-- 1 tgg tgg 1630820 Dec 14 16:06 gst.im-
$ ./gst -S --image=./gst.im tests/AnsiLoad.st
gst: Aborted
(ip 4)[] in ProcessorScheduler>>#initialize
(ip 22)Array(SequenceableCollection)>>#do:
(ip 6)[] in ProcessorScheduler>>#initialize
(ip 4)[] in BlockClosure>>#newProcessWith:
(ip 52)[] in Process>>#onBlock:at:suspend:
(ip 10) BlockClosure>>#on:do:
(ip 14)[] in Process>>#onBlock:at:suspend:
(ip 2) BlockClosure>>#ensure:
(ip 10)[] in Process>>#onBlock:at:suspend:
(ip 46)[] in BlockClosure>>#asContext:
(ip 14)BlockContext class>>#fromClosure:parent:
Aborted

Replacing gst.im with gst.im- does work, but modifies the image again.
Also notice "Global garbage collection... done" is displayed twice. (I'm using embedded libsigsegv0)

Both images are available from http://thomas.g.girard.free.fr/gst/

Yes, but where are the two executable different? Can you try compiling them in exactly the same environment, including the directory in which you build?

Also, embedded libsigsegv0 = not the Debian package?

Thanks,

I've computed md5sum for both and diffed the ouput at http://thomas.g.girard.free.fr/gst/tarball_to_patch-551.diff

Yes, I'm not using Debian libsigsegv0.

Can you redo the compilation with "-save-temps -O2 -g", and diff the interp.i and interp.s files? Thanks!

There you go:

diff -u ./smalltalk-2.95g--tarball/libgst/interp.i ./smalltalk-2.95g--patch-551/libgst/interp.i
--- ./smalltalk-2.95g--tarball/libgst/interp.i 2007-12-15 16:10:01.000000000 +0000

+++  ./smalltalk-2.95g--patch-551/libgst/interp.i        2007-12-15 16:31:01.000000000 +0000

@@ -23890,7 +23890,7 @@
   return (true);
 }
 
-int _gst_primitives_md5[4] = { 0xf7b28bad, 0xf51eaf5f, 0x2c72fbc5, 0xc7d17f62 };

+ int _gst_primitives_md5[4] = { 0xad8bb2f7, 0x5faf1ef5, 0xc5fb722c, 0x627fd1c7 };

 
 void
 _gst_init_primitives()

and

diff -u ./smalltalk-2.95g--tarball/libgst/interp.s ./smalltalk-2.95g--patch-551/libgst/interp.s
--- ./smalltalk-2.95g--tarball/libgst/interp.s 2007-12-15 16:10:22.000000000 +0000

+++  ./smalltalk-2.95g--patch-551/libgst/interp.s        2007-12-15 16:31:22.000000000 +0000

@@ -67740,10 +67740,10 @@
        .type   _gst_primitives_md5, @object
        .size   _gst_primitives_md5, 16
 _gst_primitives_md5:
-       .long   -139293779
-       .long   -182538401
-       .long   745733061
-       .long   -942571678

+        .long   -1383353609

+       .long   1605312245

+        .long   -973376980

+       .long   1652543943
        .align 4
        .type   free_lifo_context, @object
        .size   free_lifo_context, 4

Diff attached.

I've also copied compilation typescripts in http://thomas.g.girard.free.fr/gst

Hmmm... { 0xf7b28bad, 0xf51eaf5f, 0x2c72fbc5, 0xc7d17f62 } looks a lot like: { 0xad8bb2f7, 0x5faf1ef5, 0xc5fb722c, 0x627fd1c7 } read backwards...

Ok, so it's an obvious endianness issue. Less obvious is why it worked for me, and why it worked on x86-64.

Anyway here's the patch to fix this.

Funny enough, with the patch a `make distcheck' now allows me to reproduce the bug on my machine too.

It is a GC problem caused by weak objects. I'll soon commit the attached patch to fix it.