Bug #7721: test_too_long_path2(TestProcess) fails on mingw32

• Status: Closed
• Assignee: nobu (Nobuyoshi Nakada)
• Target version: -
• ruby -v: 38934

[phasis68 (Heesob Park)]

[ 6319/11317] TestProcess#test_too_long_path2Microsoft Windows [Version 6.1.7601]
Copyright (c) 2009 Microsoft Corporation. All rights reserved.

C:\work\snapshot-mg32>exit
= 8.03 s
33) Failure:
test_too_long_path2(TestProcess) [C:/work/snapshot-mg32/test/ruby/test_process.rb:1393]:
[ruby-core:34833].
[Errno::ENOENT, Errno::E2BIG] expected but nothing was raised.

This failure is also mentioned at https://bugs.ruby-lang.org/issues/7710#note-3

[phasis68 (Heesob Park)]

After some inspections, I found this failure occurs not on a very long path but on a sufficiently long path with many preloaded test sets.

On such a condition, acp_to_wstr or mbstr_to_wstr function returns NULL with the failure of malloc.

Here is a workaround patch:
diff --git a/win32.c b/win32.c.new
index 3b82766..4822f63 100644
--- a/win32.c
+++ b/win32.c.new
@@ -1126,7 +1126,7 @@ CreateChild(const WCHAR *cmd, const WCHAR *prog, SECURITY_ATTRIBUTES *psa,

 dwCreationFlags |= NORMAL_PRIORITY_CLASS;

• if (lstrlenW(cmd) > 32767) {

• if (cmd==NULL || lstrlenW(cmd) > 32767) {
  child->pid = 0; /* release the slot */
  errno = E2BIG;
  return NULL;

[jonforums (Jon Forums)]

Nice Process.spawn spelunking.

The patch also works for me on Win7 32bit with mingw-w64 4.7.2 on ruby 2.0.0dev (2013-01-22 trunk 38896) [i386-mingw32]

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r38904.
Heesob, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

win32.c: acp_to_wstr results check

• win32/win32.c (rb_w32_spawn, rb_w32_aspawn_flags): check the results
  of acp_to_wstr() which can return NULL. [ruby-core:51557] [Bug #7721]

[nobu (Nobuyoshi Nakada)]

• Tracker changed from Bug to Backport
• Project changed from Ruby master to Backport193
• Status changed from Closed to Assigned
• Assignee set to usa (Usaku NAKAMURA)

[phasis68 (Heesob Park)]

This bug is not solved with changeset r38904.

There are some warnings.

compiling win32/win32.c
win32/win32.c: In function 'rb_w32_aspawn_flags':
win32/win32.c:1296:14: warning: 'ret' may be used uninitialized in this function

win32/win32.c: In function 'rb_w32_spawn':
win32/win32.c:1187:14: warning: 'ret' may be used uninitialized in this function

win32/win32.c: In function 'rb_w32_aspawn_flags.clone.8':
win32/win32.c:1296:14: warning: 'ret' may be used uninitialized in this function

And the test still fails.

19. Failure:
    test_too_long_path2(TestProcess) [C:/work/snapshot-mg32/test/ruby/test_process.rb:1393]:
    [ruby-core:34833].
    [Errno::ENOENT, Errno::E2BIG] expected but nothing was raised.

Here is a patch:
diff --git a/win32.c b/win32.c.new
index 8b577ea..984e03b 100644
--- a/win32.c
+++ b/win32.c.new
@@ -1184,7 +1184,7 @@ rb_w32_spawn(int mode, const char *cmd, const char *prog)
const char *shell = NULL;
WCHAR *wcmd = NULL, *wshell = NULL;
int e = 0;

• rb_pid_t ret;

• rb_pid_t ret = -1;
  VALUE v = 0;
  VALUE v2 = 0;

@@ -1293,7 +1293,7 @@ rb_w32_aspawn_flags(int mode, const char *prog, char *const *argv, DWORD flags)
char *cmd, fbuf[MAXPATHLEN];
WCHAR *wcmd = NULL, *wprog = NULL;
int e = 0;

• rb_pid_t ret;

• rb_pid_t ret = -1;
  VALUE v = 0;

  if (check_spawn_mode(mode)) return -1;

[luislavena (Luis Lavena)]

• Assignee changed from usa (Usaku NAKAMURA) to nobu (Nobuyoshi Nakada)
• % Done changed from 100 to 50

I can confirm, test still fails:

http://ci.rubyinstaller.org/job/ruby-trunk-x86-test-all/703/console

test_too_long_path2(TestProcess) [C:/Users/Worker/Jenkins/workspace/ruby-trunk-x86-build/test/ruby/test_process.rb:1393]:
[ruby-core:34833].
[Errno::ENOENT, Errno::E2BIG] exception expected, not
Class:
Message: <"failed to allocate memory">
---Backtrace---
C:/Users/Worker/Jenkins/workspace/ruby-trunk-x86-build/test/ruby/test_process.rb:1393:in initialize' C:/Users/Worker/Jenkins/workspace/ruby-trunk-x86-build/test/ruby/test_process.rb:1393:in spawn'
C:/Users/Worker/Jenkins/workspace/ruby-trunk-x86-build/test/ruby/test_process.rb:1393:in `block in test_too_long_path2'

[luislavena (Luis Lavena)]

• Tracker changed from Backport to Bug
• Project changed from Backport193 to Ruby master

[phasis68 (Heesob Park)]

The NoMemoryError is raised from rb_syserr_new_str call in make_errno_exc_str function with too long path string.

Here is a patch:
diff --git a/error.c b/error.c.new
index 481c117..c26feff 100644
--- a/error.c
+++ b/error.c.new
@@ -1859,6 +1859,9 @@ make_errno_exc_str(VALUE mesg)
{
int n = errno;

+#ifdef E2BIG

• if (n == E2BIG) mesg = Qnil;
  +#endif
  errno = 0;
  if (!mesg) mesg = Qnil;
  if (n == 0) {

[nobu (Nobuyoshi Nakada)]

• Status changed from Assigned to Closed
• % Done changed from 50 to 100

This issue was solved with changeset r38934.
Heesob, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

test_process.rb: reduce command string size

• test/ruby/test_process.rb (test_spawn_too_long_path),
  (test_aspawn_too_long_path): reduce command string size until
  intended exception occurs. [ruby-core:51592] [Bug #7721]

[nobu (Nobuyoshi Nakada)]

• File 0001-error.c-defer-error-message.patch 0001-error.c-defer-error-message.patch added
• ruby -v set to 38934

NoMemoryError could occur at many places, almostly everywhere creating
a new object, adding a new instance variable, growing up a container
object, etc. It's just rare in normal cases.

r30682 reduced the sizes from 100MB to 10MB because NoMemoryError is
not intended, but test_too_long_path2 is still consuming double of
test_too_long_path.

I fix these tests instead for the time being.

Another strategy is, similar to NameError::message, keeping the given
parameters untouched and defer generating the message until to_s is
called. But it's complex considering Marshal.