Re: [Monotone-devel] server_request_sync vs pipe (uri)

[Stephen Leake]

Roland McGrath <address@hidden> writes:

> I tried a hook doing server_request_sync with a (file) uri and the mtn
> serve process went into tight loop of "got woken up for action on unknown fd".
> The fd it said was its fd for the pipe to the child 'mtn serve
> --stdio' stdin.

What OS are you running on? The --stdio code is broken for Win32 native.

I've been trying to fix it; see the n.v.m.experimental.win32_pipes
branch.

> I think I more or less understood what was going on there and kind of fixed
> it with the patch attached below.  

The patch changes `read_some' to return {-1, 0, 1} instead of bool.
Could you provide a comment explaining what each result means, and
what the rationale for this is?

> Now it works once, so the push by server_request_sync completes
> fine. But after that, the server is stuck so a client connecting
> again never gets anywhere. I think I may be out of my depth with
> poking more at this code without really trying to understand all the
> libraries (not to mention C++ features). Any help?

Perhaps you could add a test driver for this case; that would make it
easier to duplicate.

-- 
-- Stephe