[PATCH] test-commandserver: use python instead of hg as the executable

[Matt Mackall]

On Thu, 2012-06-21 at 20:41 +0200, Adrian Buehlmann wrote:
> On 2012-06-21 13:41, Adrian Buehlmann wrote:
> > On 2012-06-21 09:37, Pierre-Yves David wrote:
> >> On Thu, Jun 21, 2012 at 02:33:06AM +0200, Adrian Buehlmann wrote:
> >>> On 2012-06-21 00:42, Adrian Buehlmann wrote:
> >>>> On 2012-06-21 00:22, Matt Mackall wrote:
> >>>> diff --git a/tests/test-commandserver.py b/tests/test-commandserver.py
> >>>> --- a/tests/test-commandserver.py
> >>>> +++ b/tests/test-commandserver.py
> >>>> @@ -5,7 +5,7 @@
> >>>>      if path:
> >>>>          cmdline += ['-R', path]
> >>>>
> >>>> -    server = subprocess.Popen(cmdline, stdin=subprocess.PIPE,
> >>>> +    server = subprocess.Popen(cmdline, shell=True, stdin=subprocess.PIPE,
> >>>>                                stdout=subprocess.PIPE)
> >>>>
> >>>>      return server
> >>>>
> >>>> would be an acceptable change, then test-commandserver.py passes here
> >>>> with a file hg.cmd in C:\Users\adi\hgrepos\hg-main containing:
> >>>
> >>> On Ubuntu Linux, test-commandserver.py crashes here with that patch:
> >>>
> >>> +Traceback (most recent call last):
> >>> +  File "/home/adi/hgrepos/hg-main/tests/test-commandserver.py", line 242, in <module>
> >>> +    check(hellomessage)
> >>> +  File "/home/adi/hgrepos/hg-main/tests/test-commandserver.py", line 63, in check
> >>> +    return func(server)
> >>> +  File "/home/adi/hgrepos/hg-main/tests/test-commandserver.py", line 72, in hellomessage
> >>> +    ch, data = readchannel(server)
> >>> +  File "/home/adi/hgrepos/hg-main/tests/test-commandserver.py", line 26, in readchannel
> >>> +    return channel, server.stdout.read(length)
> >>> +MemoryError
> >>
> >> They are tons of possible cause for a MemoryError, the most common one being
> >> "Firefox is eating your ram with a Ladle".
> >>
> >> I do not see why switching to shell=True should be the main suspect for such
> >> error. I just tried on linux (debian squeeze) and it works great.
> > 
> > Ok. Thanks for testing.
> > 
> > It can't be related to Firefox here, as I don't use that. This is a
> > 32-bit Ubuntu in a VirtualBox VM here, which I only use for mercuial
> > testing. The most demanding application I use there is probably
> > TortoiseHg, but even if that's running, there is plenty of free RAM left
> > (~85% free).
> > 
> > This is a 32-bit Ubuntu 11.04 (natty), with Kernel Linux
> > 2.6.38-15-generic, 1.4 GiB RAM, Python 2.7.1+ (r271:86832, Apr 11 2011,
> > 18:05:24).
> > 
> > The MemoryError consistently shows up with the patch applied, and not
> > without the patch applied.
> > 
> > This is the first time I have such problem with that VM.
> > 
> > So I guess I'm tripping over some bug which might be unrelated to
> > mercurial, but that usage triggers the bug. :(
> > 
> > Perhaps I should ditch that VM and install a newer Ubuntu.
> 
> So.. I have now switched into completely different Ubuntu install in a
> different VM. This time, it's a 64-bit VM.

That's very odd. I get weird behavior, but no MemoryError. I suspect
what's happening is:

- subprocess is doing something dumb with argument lists when shell mode
is used
- hg gets launched with no args
- hg outputs some banner that gets interpreted as protocol
- some ASCII characters get interpreted as "here comes a huge read"
- test-commandserver tries to allocate a big buffer
- MemoryError on small boxes

 This works here:

$ hg diff
diff -r 21e18c608b68 tests/test-commandserver.py
--- a/tests/test-commandserver.py	Thu Jun 14 11:43:48 2012 +0200
+++ b/tests/test-commandserver.py	Thu Jun 21 14:53:42 2012 -0500
@@ -2,11 +2,12 @@
 
 def connect(path=None):
     cmdline = ['hg', 'serve', '--cmdserver', 'pipe']
+    cmdline = ' '.join(cmdline)
     if path:
         cmdline += ['-R', path]
 
     server = subprocess.Popen(cmdline, stdin=subprocess.PIPE,
-                              stdout=subprocess.PIPE)
+                              stdout=subprocess.PIPE, shell=True)
 
     return server
 
-- 
Mathematics is the supreme nostalgia of our time.