[Bioc-devel] AnnotationHub: cleanup

[Kasper Daniel Hansen]

On Tue, Sep 15, 2015 at 12:25 AM, Morgan, Martin <
Martin.Morgan at roswellpark.org> wrote:

> Hi Kasper -- we'll try to act on these, but some comments / looking for
> clarification...
>
> > -----Original Message-----
> > From: Bioc-devel [mailto:bioc-devel-bounces at r-project.org] On Behalf Of
> > Kasper Daniel Hansen
> > Sent: Monday, September 14, 2015 10:45 PM
> > To: bioc-devel at r-project.org
> > Subject: [Bioc-devel] AnnotationHub: cleanup
> >
> > I currently have the `pleasure` of dealing with students who have
> problems
> > with installing AnnotationHub and/or downloading resources.  Here are
> some
> > comments including some possible bug reports.
>
> I hope this is on the whole a positive experience, and we'll do what we
> can to make it better.
>

Well, I love the package and I love it even more having prepared material
on it.  And the people who complain is of course enriched for people who
have problems - no way to know if it just works for most people.

And of course right now it is more troublesome since I prepared the class
using R-3.2.1 and then 3.2.2 was released just before we started and had
the http -> https change which is an obvious suspect when people have
download problems :)

 > 1) I think it is extremely dangerous that `cache(ahub)` starts by asking
to

> > download all resources!  May I suggest this only happens with a specific
> > setting like `cache(ahub, download=TRUE)` or something similar.
>
> >
> > 2) `cache(ahub)` deletes all cached information, except the sqlite
> database.
> > Could we get a way to remove everything?
> >
> > 3) While I can understand the difference between cache and hubCache, I
> > would suggest that hubCache(ahub) = NULL removes all cached material
> > included the sqlite database.
>
> For each of the above the envisioned use case was that  'hub' is a subset,
> eg.,
>
>   subhub = query(hub, c("homo", "ensembl", "81"))
>
> and the user wanted to manipulate all records in the sub-hub.
> cache(subhub) asks about the 'really download" if the size of the (sub)hub
> is greater than hubOption("MAX_DOWNLOADS"), which by default is 10; it
> seems like asking is the same as requiring an argument? fileName(subhub)
> may be closer to what you're looking for...? the path to the file name, or
> NA if It is not in the cache.
>
> For cache(subhub) = NULL it wouldn't make sense to delete 5 resources AND
> the sqlite file for the entire hub.
>
> The sqlite file can be discovered with dbfile(hub) / dbfile(subhub), and
> removed with file.remove(dbfile(subhub))). In some ways it wasn't
> envisioned that this manual manipulation would be a common use case (!).


Ok.  Let me perhaps rephrase my wish list
1) some easy way to reset the entire cache issue, with emphasis on easy.
This is most likely to be used by beginners.  Who it's done, I don't care
to much about.  And I suggest a heading in ?AnnotationHub called something
like "Flushing the cache" or something.
2) It seems natural that there is a way (for problem reporting) to report
which resources are cached, which is (again) easy and does not involve
download.  I don't care if it is cache() or some other name.

> 4) It seems that AnnotationHub in the release version of Bioconductor
> > defaults to using https://.  Wasn't full support for https://
> introduced in R
> > 3.2.2; if so, it seems to be a critical bug that it is using https://
>
> AnnotationHub uses httr::GET and ultimately curl::curl_fetch_disk rather
> than native R support, so what R does is not directly relevant. From ?curl
>
>      Drop-in replacement for base 'url' that supports https, ftps,
>      gzip, deflate, etc. Default behavior is identical to 'url', but
>      request can be fully configured by passing a custom 'handle'.
>
> So I wonder what the actual problem is?
>

Interesting.  Well, at least one user is behind a proxy and uses the tips
in ?download.file to set a proxy server.  Perhaps that doesn't work with
httr?  I don't know.  But there are more than one person with problems.

> 5) Perhaps it should be considered that the default hubCache path is
> > versioned, perhaps with Bioc version, perhaps with something else.  This
> > might cause problems for people running multiple versions of R.
>
> The data base is supposed to handle versioning, so if you've populated the
> cache with Bioc 3.2 and are now accessing the cache with Bioc 3.1, only the
> 3.1 resources are visible. The hope was to avoid multiple copies of these
> possibly large resources.


That sounds pretty nifty.. I was thinking re-design of the database issues.


> 6) I strongly suggest that the output printed when retrieving an

> > AnnotationHub resource includes the download url.
>
> Ok something that's easy to do! Sometimes this will be cryptic (when the
> resource is cached in the AnnotationHub server, rather than being retrieved
> from the original source)


Perhaps it should just say "loading from cache"


> > 7) If you run AnnotationHub without having GenomicRanges / rtracklayer
> > installed, it downloads the resource and then pangs out with an error.
> To me
> > it seems more natural to pang out with an error immediately, especially
> since
> > when it works, it appears from message printing that loading the library
> > happens prior to download.
>
> I guess by 'run AnnotationHub' you mean retrieve a specific resource?
>
> The import recipes generally start by require()ing the necessary
> libraries. I spotted a couple of recipes that didn't follow this convention
> (for 2bit and chain file resources from rtracklayer; none that involved
> GenomicRanges). Are there specific examples?
>

As a test case I got a Windows virtual machine up and running, total clean,
 and just did biocLite("AnnotationHub").  Then I picked two random
resources and tried to download them; one was a UCSC chain file and I don't
know the other one.  In both cases I totally got a decent error message,
which I can fully understand.  But looking at it with beginner eyes, I just
thought it was weird that the error on missing a library happened after
download.  It's not a bit deal, but if you don't know what you're doing you
might get confused.

Best,
Kasper

	[[alternative HTML version deleted]]