Re: [gpsd-dev] tests hanging on netsbd-6 i386

[Eric S. Raymond]

Greg Troxel <address@hidden>: 
> With the assert fix, that test hangs.  If I kill gpsd, the next test
> starts and hangs.  (There's a bug; abend should result in test failure.)

Can you bisect to identify the commit where the falure entered?  Do you know
what function it is hanging in?

> I went to set buzzkill=no to disable the new code, and found that it was
> gone.

That's correct; I have removed the adaptive-delay code.  Chunking the low-level
I/O works better and is less intrustive; it even improves latency for
time service.

> In general, I think the cope-with-buggy-select stuff is impacting too
> much on systems that don't have buggy select.  How prevalent is this
> really?  Is the issue that select returns that a fd is readable when it
> isn't?  Can this be detected at runtime?   Can the code just sleep for
> 100ms after getting a zero-length read when select said it had data,
> avoiding impacting non-broken platforms?

Sleeping for a fixed interval is exactly what the second version of the
buzzkill code did.  That seemed to produce hangs.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>

signature.asc
Description: Digital signature