[PATCH] don't silently ignore unknown file types

[Mike Bonnet]

The attached patch causes Maven builds to fail if they produce a file 
type that Koji doesn't know about.  This is preferable to silently 
ignoring them.

[Mike McLean]

On 11/06/2015 04:49 PM, Mike Bonnet wrote:
> The attached patch causes Maven builds to fail if they produce a file
> type that Koji doesn't know about.  This is preferable to silently
> ignoring them.

I had sort of assumed the code was already doing this. It does for
images, and for manual imports (the latter because we run
get_archive_type with strict on).

So, we're going from silently ignoring a condition to failing on it. Do
we have any sense of how often this happens?

Thinking further down the road, this really seems like the wrong place
to have this check. I'm not sure the hub should really care about this.
Also, seems silly for the builder to bother uploading content that's
going to be rejected.   ...maybe not something we're going to manage to
fix in 1.x though

While we're at it should we also make the same change in completeWinBuild?

[Mike Bonnet]

On 11/6/15 2:46 PM, Mike McLean wrote:
> On 11/06/2015 04:49 PM, Mike Bonnet wrote:
> > The attached patch causes Maven builds to fail if they produce a file
> > type that Koji doesn't know about.  This is preferable to silently
> > ignoring them.
>
> I had sort of assumed the code was already doing this. It does for
> images, and for manual imports (the latter because we run
> get_archive_type with strict on).
>
> So, we're going from silently ignoring a condition to failing on it. Do
> we have any sense of how often this happens?

It has happened a couple times recently, with undesirable results.  The 
affected teams are the ones requesting this change.
> Thinking further down the road, this really seems like the wrong place
> to have this check. I'm not sure the hub should really care about this.
> Also, seems silly for the builder to bother uploading content that's
> going to be rejected.   ...maybe not something we're going to manage to
> fix in 1.x though

There are some good reasons to upload content which we don't want in the 
database.  For example, for Maven builds we don't import 
maven-metadata.xml (it's regen'ed on a per-repo basis because it 
aggregates information from multiple builds) but we want it in the task 
output directory because that directory is then a valid Maven repo, 
which we point subsequent tasks at in a maven-chain build.
> While we're at it should we also make the same change in completeWinBuild?

Agreed, updated patch attached.
> koji-devel mailing list
> [e-mail]
> https://lists.fedorahosted.org/mailman/listinfo/koji-devel

[Mike McLean]

On 11/06/2015 06:49 PM, Mike Bonnet wrote:
> On 11/6/15 2:46 PM, Mike McLean wrote:
> > On 11/06/2015 04:49 PM, Mike Bonnet wrote:
> > > The attached patch causes Maven builds to fail if they produce a file
> > > type that Koji doesn't know about.  This is preferable to silently
> > > ignoring them.
> >
> > I had sort of assumed the code was already doing this. It does for
> > images, and for manual imports (the latter because we run
> > get_archive_type with strict on).
> >
> > So, we're going from silently ignoring a condition to failing on it. Do
> > we have any sense of how often this happens?
>
> It has happened a couple times recently, with undesirable results.  The
> affected teams are the ones requesting this change.
> > Thinking further down the road, this really seems like the wrong place
> > to have this check. I'm not sure the hub should really care about this.
> > Also, seems silly for the builder to bother uploading content that's
> > going to be rejected.   ...maybe not something we're going to manage to
> > fix in 1.x though
>
> There are some good reasons to upload content which we don't want in the
> database.  For example, for Maven builds we don't import
> maven-metadata.xml (it's regen'ed on a per-repo basis because it
> aggregates information from multiple builds) but we want it in the task
> output directory because that directory is then a valid Maven repo,
> which we point subsequent tasks at in a maven-chain build.

Ok, but seems like there should be some clearer distinction between the
different types of uploaded content and that the hub should be less
involved in making this call. Basically, I think the error condition
this is addressing should be dealt with before this stage.

Still, this is a reasonable start. We can haggle about the archivetypes
table in 2.0 ;)

Pulling in
> > While we're at it should we also make the same change in
> > completeWinBuild?
>
> Agreed, updated patch attached.
> > koji-devel mailing list
> > [e-mail]
> > https://lists.fedorahosted.org/mailman/listinfo/koji-devel
>
> koji-devel mailing list
> [e-mail]
> https://lists.fedorahosted.org/mailman/listinfo/koji-devel