Think I found a bunch of the makehistory problem

[Russ Allbery]

Jeffrey M Vinocur <jeff at litech.org> writes:
> On Sun, 10 Mar 2002, Russ Allbery wrote:

>> You may want to try poking at overview with tdx-util too to see if you
>> can figure out why it might be so slow.  Just to see what all is in the
>> current overview files.

> Sure, will do.  (Is there any documentation for it?)

Nope, not yet.  -d dumps the group.index file, -i gives you the data for a
particular group, -g dumps the .IDX file for that group, and -o dumps the
actual overview data for that group.  If you want only a particular
article's overview data, use -n number.

>> Hm.  I wonder if write is returning a partial write to disk.  (No error
>> message, but not writing out all of the data.)  That would suck; that
>> would mean that we'd have to put loops around all those write calls.

> Ohh.  Ugh.  (Well, we might as well just put a wrapper in lib/ someplace.)

We already have an xwrite wrapper, but pwrite is a bit trickier since you
have to adjust the offset on the next write.  I guess we need an xpwrite
wrapper, at least as a try.

> I wonder if we should try to zero errno after we ignore an error, so at
> least we'd know when we're reporting bogus values.  Probably too hard to
> track down those cases, though, even within our own code.

The thing is, libc is allowed to arbitrarily fiddle with errno as long as
it doesn't return an error, and you see that all the time.  Particularly
when calling functions like gethostbyname that do a bunch of random work
behind the scenes.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.