suboptimal fetch behavior [was: No valid mirrors for mysql56?

[John Marino]

On 7/18/2013 01:29, John Marshall wrote:
>> Seems odd that the ports system wouldn't catch this and delete
>> the bad original before trying to re-download or forcing an overwrite
>> on download.
> 
> So, the tarball had changed since the last time you downloaded it (using
> an earlier version of the port).  The ports system did catch it and
> complain that your downloaded tarball didn't match the (new) expected
> tarball for the same release, but the ports system didn't assume
> responsibility for deleting the existing file.

This behavior has personally burned me numerous times.  As I am often
building the entire tree, this behavior not only causes false positives
on bad builds, but it prevents all the ports that depend on it from
building as well which hurts builds meant to produce official packages.

I think the behavior is sub-optimal and needs to be improved.  Something
along the lines of downloading it to a temporary file, and if that file
matches the hash then replacing the current distfile with new one.

Yeah, I definitely think this should be looked at.

The other thing: This doesn't affect pkgsrc often because when a package
gets rerolled they immediately switch to a unique SUBDIR scheme so that
the new distfile never clashes with the old one.  That's a policy change
on how to handle re-rolls but it does work.

John