3.0.0 JIT segfaults on 64-bit Cygwin

Guile 2.9.9, like .8 and .7, does not build on Cygwin (64 bit). Configure

runs without error, but make crashes with this (truncated to just the tail):

Making all in bootstrap

make[2]: Entering directory

'/cygdrive/c/Users/rr828893/Downloads/guile-2.9.9/bootstrap'

BOOTSTRAP GUILEC ice-9/eval.go

BOOTSTRAP GUILEC ice-9/psyntax-pp.go

BOOTSTRAP GUILEC language/cps/intmap.go

BOOTSTRAP GUILEC language/cps/intset.go

BOOTSTRAP GUILEC language/cps/graphs.go

BOOTSTRAP GUILEC ice-9/vlist.go

BOOTSTRAP GUILEC srfi/srfi-1.go

/bin/sh: line 6: 4294 Segmentation fault (core dumped)

GUILE_AUTO_COMPILE=0 ../meta/build-env guild compile

--target="x86_64-unknown-cygwin" -O1 -Oresolve-primitives -L

"/home/rr828893/Downloads/guile-2.9.9/module" -L

"/home/rr828893/Downloads/guile-2.9.9/guile-readline" -o "srfi/srfi-1.go"

"../module/srfi/srfi-1.scm"

make[2]: *** [Makefile:1930: srfi/srfi-1.go] Error 139

make[2]: Leaving directory

'/cygdrive/c/Users/rr828893/Downloads/guile-2.9.9/bootstrap'

make[1]: *** [Makefile:1849: all-recursive] Error 1

make[1]: Leaving directory

'/cygdrive/c/Users/rr828893/Downloads/guile-2.9.9'

make: *** [Makefile:1735: all] Error 2

All previous problems (which were easy to work around) have gone away in

this release, which is progress, but it doesn't get me past Guile 2.2.

John Cowan http://vrici.lojban.org/~cowan cowan@ccil.org

Your worships will perhaps be thinking that it is an easy thing

to blow up a dog? [Or] to write a book?

--Don Quixote, Introduction

Okay, with GUILE_JIT_THRESHOLD set to -1 in the environment, I can build

Guile under Cygwin. There are two test failures which probably reflect

differences between newlib and glibc:

ERROR: time.test: strptime: GNU %s format: strftime fr_FR.utf8 - arguments:

((system-error "strptime" "~A" ("Invalid argument") (22)))

ERROR: time.test: strptime: GNU %s format: strftime fr_FR.iso88591 -

arguments: ((system-error "strptime" "~A" ("Invalid argument") (22)))

And that's that: Cygwin can support Guile 3.0 without JIT. It might be a

good idea to force this variable on in "configure" when building under

Cygwin.

retitle 39118 3.0.0 JIT segfaults on 64-bit Cygwin

quit

Hi John,

John Cowan <cowan@ccil.org> skribis:

Could you try building 3.0.0 with JIT enabled and grab a backtrace?

Thanks in advance!

Ludo’.

Yes, gladly, but I don't know how to get one in this context. Do I need to

add some flags to the Makefile, and if so, where? (It's a twisty maze of

passages, all different.) . Note that this *is* a build with JIT enabled;

when I disable it using the env variable, there are no errors and 3.0.0

works fine.

Also, it may take some time, as I have to rebuild my Windows system.

On Mon, Jan 20, 2020 at 11:38:35AM -0500, John Cowan wrote:

I also tried building Guile 3.0.0 on Cygwin 3.1.x. The failure comes from

trying to parse compiled .go files.

The last time that I had this sort of problem, it was because the

O_BINARY flag was dropped or missing when writing .go files, leading

to CR+LF characters in the compiled files. And I diagnosed it by

byte-comparing Linux-compiled .go files with Cygwin-compiled .go

files, and by looking for CR+LF combinations in the compiled .go

files.

I don't know if that is what is happening here, but, I'll check that

next time I have a chance.

Thanks,

Michael

Hello,

John Cowan <cowan@ccil.org> skribis:

You would unpack, configure, and build like you did before (with JIT

enabled, so as to reproduce the crash), but before that you’d run

“ulimit -c unlimited” in that shell to make sure there’s a core dumped

when it crashes.

Once it has crashed, locate the ‘core’ file (or ‘core.*’), and run, say:

gdb libguile/.libs/guile bootstrap/core

Then from the GDB prompt:

thread apply all bt

TIA,

Ludo’.

On Tue, Jan 21, 2020 at 10:01:58AM +0100, Ludovic Courtès wrote:

This won't work on cygwin. If you want a core dump, you should use the

dumper tool, as described here

Or you can set error_start to gdb to get an interactive gdb session on error.

s

Thanks. Unfortunately, the standard recipe for making core dumps on Mac

(put "limit core unlimited" into /etc/launchd.conf and reboot, make sure

/cores is writable, set ulimit -c unlimited) seem to actually enable them

on MacOS Catalina (10.15.2). I have tested with SIGQUIT and SIGSEGV on

running processes and no dumps appear in /cores.

On Tue, Jan 21, 2020 at 4:02 AM Ludovic Courtès <ludo@gnu.org> wrote:

I'm no longer talking about Cygwin (which builds fine without JIT). I'm

now talking about MacOS Catalina, which needs a core dump to debug, but on

which nobody seems to know how to enable core dumps.

On Tue, Jan 21, 2020 at 1:41 PM szgyg <szgyg@ludens.elte.hu> wrote:

Hi,

John Cowan <cowan@ccil.org> skribis:

This bug report is about Cygwin, not macOS, right? :-)

Ludo’.

Both Cygwin and MacOS crash in pretty much the same way. By disabling the

JIT, I was able to get the Cygwin build to run to completion. On MacOS

with --disable-jit, however, I am now getting an entirely new failure:

CC readline.lo

readline.c:432:7: warning: implicitly declaring library function 'strncmp'

with type 'int (const char *, const char *,

unsigned long)' [-Wimplicit-function-declaration]

if (strncmp (rl_get_keymap_name (rl_get_keymap ()), "vi", 2))

^

readline.c:432:7: note: include the header <string.h> or explicitly provide

a declaration for 'strncmp'

readline.c:432:16: warning: implicit declaration of function

'rl_get_keymap_name' is invalid in C99

[-Wimplicit-function-declaration]

if (strncmp (rl_get_keymap_name (rl_get_keymap ()), "vi", 2))

^

readline.c:432:16: warning: incompatible integer to pointer conversion

passing 'int' to parameter of type 'const char *'

[-Wint-conversion]

if (strncmp (rl_get_keymap_name (rl_get_keymap ()), "vi", 2))

^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

3 warnings generated.

CCLD guile-readline.la

Undefined symbols for architecture x86_64:

"_rl_get_keymap_name", referenced from:

_scm_init_readline in readline.o

ld: symbol(s) not found for architecture x86_64

clang: error: linker command failed with exit code 1 (use -v to see

invocation)

On Thu, Jan 23, 2020 at 3:35 PM Ludovic Courtès <ludo@gnu.org> wrote:

Pretty sure that the missing readline symbol is because the macos

readline is being used/found instead of GNU readline.

-Dale

-----------------------------------------From: "John Cowan"

To: "Ludovic Courtès"

Cc: 39118@debbugs.gnu.org, guile-devel@gnu.org

Sent: Friday January 24 2020 9:36:59AM

Both Cygwin and MacOS crash in pretty much the same way. By disabling

the JIT, I was able to get the Cygwin build to run to completion. On

MacOS with --disable-jit, however, I am now getting an entirely new

failure:

CC readline.lo

readline.c:432:7: warning: implicitly declaring library function

'strncmp' with type 'int (const char *, const char *,

unsigned long)' [-Wimplicit-function-declaration]

if (strncmp (rl_get_keymap_name (rl_get_keymap ()), "vi", 2))

^

readline.c:432:7: note: include the header or explicitly provide a

declaration for 'strncmp'

readline.c:432:16: warning: implicit declaration of function

'rl_get_keymap_name' is invalid in C99

[-Wimplicit-function-declaration]

if (strncmp (rl_get_keymap_name (rl_get_keymap ()), "vi", 2))

^

readline.c:432:16: warning: incompatible integer to pointer conversion

passing 'int' to parameter of type 'const char *'

[-Wint-conversion]

if (strncmp (rl_get_keymap_name (rl_get_keymap ()), "vi", 2))

^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

3 warnings generated.

CCLD guile-readline.la [1]

Undefined symbols for architecture x86_64:

"_rl_get_keymap_name", referenced from:

_scm_init_readline in readline.o

ld: symbol(s) not found for architecture x86_64

clang: error: linker command failed with exit code 1 (use -v to see

invocation)

On Thu, Jan 23, 2020 at 3:35 PM Ludovic Courtès wrote:

Hi,

John Cowan skribis:

> Thanks. Unfortunately, the standard recipe for making core dumps on

Mac

This bug report is about Cygwin, not macOS, right? :-)

Ludo’.

Links:

------

[1] http://guile-readline.la

[2] mailto:ludo@gnu.org

[3] mailto:cowan@ccil.org

John Cowan <cowan@ccil.org> skribis:

That I understand. However, I was asking for the backtrace of the crash

on Cygwin when JIT is enabled. Could you grab it?

Thanks in advance,

Ludo’.

On Sat, Jan 25, 2020 at 8:51 AM Ludovic Courtès <ludo@gnu.org> wrote:

1. The wisdom of the Internet has not been able to figure out how to

generate a core dump on MacOS 10.15.2 (Catalina). The usual set of

enabling steps can be performed without error, but still no core dump.

2. Until today I believed that there was no way to generate a Cygwin core

dump. I know now that there is, but I may not be able to test it until

Monday. I'll let you know, and hopefully that will provide insight into

the MacOS problem as well.

3. I will try to work further on the MacOS libffi problem (which surfaces

when you do --disable-jit to bypass the above problem) to convince MacOS to

use GNU libffi rather than the native one. It probably has to do with

pkg-config, which I barely understand.

"All problems are config problems."

John Cowan http://vrici.lojban.org/~cowan cowan@ccil.org

We are lost, lost. No name, no business, no Precious, nothing. Only empty.

Only hungry: yes, we are hungry. A few little fishes, nassty bony little

fishes, for a poor creature, and they say death. So wise they are; so just,

so very just. --Gollum

Aaaand... Cygwin doesn't do core dumps. Under the skin it's WIndows, after

all. This is what I get when I specify ulimit -c unlimited and rebuild:

Exception: STATUS_ACCESS_VIOLATION at rip=0055A8B1B25

rax=0000000000000000 rbx=FFFFFFFFFFFFFF90 rcx=FFFFFFFFFFFFFF90

rdx=000000000034964A rsi=000007000084ECC0 rdi=FFFFFFFFFFFFFF90

r8 =000007000084ECC0 r9 =0000000000000002 r10=0000000100000000

r11=000000055A86B190 r12=0000000000000002 r13=000000055A931EA0

r14=000006FFFFFEF840 r15=0000000000000000

rbp=000000000034964A rsp=00000000FFFFBDA0

program=C:\Users\rr828893\Downloads\guile-3.0.0\libguile\.libs\guile.exe,

pid 62833, thread main

cs=0033 ds=002B es=002B fs=0053 gs=002B ss=002B

I can't imagine what you can make of that.

On Sat, Jan 25, 2020 at 10:54 AM John Cowan <cowan@ccil.org> wrote:

On Fri, Jan 31, 2020 at 09:23:19AM -0500, John Cowan wrote:

Please see my previous mail on how to get a real core dump on cygwin

#v+

BOOTSTRAP GUILEC ice-9/eval.go

BOOTSTRAP GUILEC ice-9/psyntax-pp.go

BOOTSTRAP GUILEC language/cps/intmap.go

BOOTSTRAP GUILEC language/cps/intset.go

BOOTSTRAP GUILEC language/cps/graphs.go

BOOTSTRAP GUILEC ice-9/vlist.go

BOOTSTRAP GUILEC srfi/srfi-1.go

Thread 1 received signal SIGSEGV, Segmentation fault.

[Switching to Thread 7444.0x2640]

0x000000055a8b1b25 in scm_to_uint64 (val=val@entry=0xffffffffffffff90) at ../../guile-3.0.0/libguile/conv-uinteger.i.c:44

44 else if (SCM_BIGP (val))

(gdb) bt

#0 0x000000055a8b1b25 in scm_to_uint64 (val=val@entry=0xffffffffffffff90) at ../../guile-3.0.0/libguile/conv-uinteger.i.c:44

#1 0x000000055a86b1ea in scm_bytevector_copy_x (source=0x700000948620, source_start=0x34964a, target=0x700000907600, target_start=0x2, len=0xffffffffffffff90)

at ../../guile-3.0.0/libguile/bytevectors.c:604

#2 0x00006ffffe743866 in ?? ()

Backtrace stopped: previous frame inner to this frame (corrupt stack?)

BOOTSTRAP GUILEC language/tree-il.go

(gdb) bt

#0 0x000000055a8b1b25 in scm_to_uint64 (val=val@entry=0xffffffffffffff90)

at ../../guile-3.0.0/libguile/conv-uinteger.i.c:44

#1 0x000000055a86b1ea in scm_bytevector_copy_x (source=0x70000055f160, source_start=0x34964a, target=0x700000808c90,

target_start=0x2, len=0xffffffffffffff90) at ../../guile-3.0.0/libguile/bytevectors.c:604

#2 0x00006ffffe73f936 in ?? ()

Backtrace stopped: previous frame inner to this frame (corrupt stack?)

BOOTSTRAP GUILEC language/tree-il/analyze.go

(gdb) bt

#0 0x000000055a8b1b25 in scm_to_uint64 (val=val@entry=0xffffffffffffff90)

at ../../guile-3.0.0/libguile/conv-uinteger.i.c:44

#1 0x000000055a86b1ea in scm_bytevector_copy_x (source=0x700000645160, source_start=0x34964a, target=0x7000008012d0,

target_start=0x2, len=0xffffffffffffff90) at ../../guile-3.0.0/libguile/bytevectors.c:604

#2 0x00006ffffe753fc6 in ?? ()

Backtrace stopped: previous frame inner to this frame (corrupt stack?)

^C

#v-

s

On Mon, Feb 3, 2020 at 5:11 PM szgyg <szgyg@ludens.elte.hu> wrote:

On Fri, Jan 31, 2020 at 09:23:19AM -0500, John Cowan wrote:

Okay, I looked at that page. However, Cygwin's dumper requires you to know

the Windows PID of the process to dump. Clearly it is intended for a

long-running process such as a server process, which you can force to core

dump, as if by "/bin/kill -SIGSEGV pid"; it is not suitable for a process

that gets a segmentation violation for internal reasons. In any case, when

building, I have no idea of the pid of the process which is dumping; it

starts up and then dumps immediately.

John Cowan http://vrici.lojban.org/~cowan cowan@ccil.org

The Penguin shall hunt and devour all that is crufty, gnarly and

bogacious; all code which wriggles like spaghetti, or is infested with

blighting creatures, or is bound by grave and perilous Licences shall it

capture. And in capturing shall it replicate, and in replicating shall

it document, and in documentation shall it bring freedom, serenity and

most cool froodiness to the earth and all who code therein. --Gospel of Tux

On Wed, Feb 05, 2020 at 04:11:04PM -0500, John Cowan wrote:

| One common way to use dumper is to plug it into cygwin's Just-In-Time

| debugging facility by adding

| error_start=x:\path\to\dumper.exe

| to the CYGWIN environment variable. Please note that x:\path\to\dumper.exe

| is Windows-style and not cygwin path. If error_start is set this way, then

| dumper will be started whenever some program encounters a fatal error.

s

On Mon 20 Jan 2020 18:22, Mike Gran <spk121@yahoo.com> writes:

Given that John said that compilation went fine with

GUILE_JIT_THRESHOLD=-1, I think perhaps this problem may have been fixed

in the past. My suspicions are that this issue is an ABI issue with

lightening that could perhaps be reproduced by:

git co https://gitlab.com/wingo/lightening

cd lightening

make -C tests test-native

Of course any additional confirmation is useful and welcome!

Cheers,

Andy

On 2/6/20, Andy Wingo <wingo@pobox.com> wrote:

I haven't been able to get guile to compile under Cygwin (just a

compilation error I haven't had time to track down), but I was able to

quickly try the above. I get:

Testing: test-native-call_10

call_10.c:9: assertion failed: e == 4

/bin/sh: line 1: 7063 Aborted (core dumped) ./$test

make: *** [Makefile:31: test-native] Error 134

Despite what it says about a core dump, I find no such thing. Just a

file with the same name as the executable suffixed with ".stackdump".

(I did attempt to configure the Cygwin dumper before running the

tests.) Unless somebody suggests otherwise, I think the error message

is more useful.

--

Charles

On 2/6/20, Charles Stanhope <charles@stanho.pe> wrote:

Andy, I don't know if you'd want to continue this here or on

lightening's gitlab page, but I looked into this a little bit a few

minutes here and there this past weeek. The x86 "fast-call" calling

convention used on Windows x64[0] and shared by Cygwin[1] requires

that the caller reserve 32 bytes of memory on the stack for the callee

to spill the register parameters (even if the callee takes fewer than

four parameters). I think lightening is currently missing that for the

x64 case for Cygwin.

To test the idea, I made a small modification (patch attached) that is

*not* intended as a solution as it doesn't work for the general case,

but it does allow the tests to pass on Cygwin 64.

[0] https://docs.microsoft.com/en-us/cpp/build/x64-calling-convention?view=vs-2019

[1] https://cygwin.com/cygwin-ug-net/programming.html#gcc-64

--

Charles

On Fri, Feb 14, 2020 at 09:46:04AM -0800, Charles Stanhope wrote:

I can confirm that Charles's patch, plus another one line patch

to define CPU_SETSIZE, is enough to get Guile 3.0.x to build and run

on my box. All tests pass except strptime in French, and the absence

of crypt. This is a 64-bit build.

-Mike Gran

Excellent, and thank you all! I've been WIndowsless for a few weeks, but

that should change again soon.

On Sun, Feb 16, 2020 at 6:23 PM Mike Gran <spk121@yahoo.com> wrote:

On 2/16/20, Mike Gran <spk121@yahoo.com> wrote:

Mike, thanks for going further with the Guile build. The CPU_SETSIZE

issue was what was hanging me up from compiling before Andy's comment

got me to look at lightening. I assumed I had some configuration,

package, or compiler issue. Good to know there's a simple fix.

Just a further warning to anyone watching, that patch I posted is a

real hack job just to test my theory of the cause of the segfault. I

would expect it to fail when you have fewer than four arguments in a

JITed function call. I wouldn't try doing much else with that Guile

build besides run the tests. :)

--

Charles

On 2/16/20, Charles Stanhope <charles@stanho.pe> wrote:

I had a little bit more time to look into the lightening

implementation last night. I've attached a patch that is less horrible

and more correct than my previous one. It reserves the stack space

regardless of the number of parameters and appears to work. But I'm

new to the lightening code base, so I'm not convinced it is the

correct solution. It's just the solution I was left with after my time

ran out. I wanted to post this patch as a replacement to the prior one

in case people did want to do more testing with Guile 3.0 on Cygwin

x64.

With that, I will let more experienced people come up with the

appropriate solution. Happy hacking, everybody!

--

Charles

Aah, you all are amazing -- thank you!! Applied and merged.

Cheers,

Andy

On Mon 17 Feb 2020 20:27, Charles Stanhope <charles@stanho.pe> writes: