https://bz.mercurial-scm.org/show_bug.cgi?id=5516
Bug ID: 5516
Summary: hglib doesn't reset the command server after an
unexpected prompt
Product: Mercurial
Version: 4.1.1
Hardware: PC
OS: Windows
Status: UNCONFIRMED
Severity: bug
Priority: urgent
Component: hglib
Assignee: bugzi...@mercurial-scm.org
Reporter: gabor.stefa...@nng.com
CC: mercurial-devel@mercurial-scm.org
When hglib encounters an unexpected prompt from the command server, it raises a
ResponseError exception. However, it fails to reset/resynchronize with the
command server. As a result, if you catch that exception, and try to run
another command, it will behave in unpredictable ways, up to and including a
hard lockup that doesn't even respond to Ctrl+C.
As an example, load the following in-process hook:
from mercurial import util
def hook(ui, repo, *args, **kwargs):
try:
repo.ui.promptchoice("dummy $$ &OK", 0)
except util.Abort:
pass
return False
as a "pre-pull" hook. Then, execute the following code:
client = hglib.open('.')
try:
client.pull()
except hglib.error.ResponseError:
pass
newtip = client.log(revrange='tip')
On my machine, this locks up in client.log inside a read(), and the command
server process itself leaks several gigabytes of memory. The only way out is to
manually kill the command server in Task Manager. It appears that hglib never
resets the command server to a "command waiting" state, and so the log command
is actually interpreted as a response to the prompt.
We should either reset the command server before raising the exception, or flag
the client as dead and deny further commands on it.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
Mercurial-devel mailing list
Mercurial-devel@mercurial-scm.org
https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel
