[Bioc-devel] package review help
Martin Morgan
martin.morgan at roswellpark.org
Sat Oct 1 11:44:08 CEST 2016
On 09/30/2016 09:53 PM, Vincent Carey wrote:
> i would have comments on StarTrek. if you would like me to be reviewer
> of that i can do it.
>
> a query on policy
>
> how do you propose to handle needlessly repeated code?
It can be very helpful to indicate sub-optimal approaches to coding to
the authors, and certainly in this case.
There is the risk of an analog of 'premature optimization', where the
code is implemented inefficiently / poorly but it does not matter for
the overall functioning of the package; it can be hard to know from a
more-or-less superficial look at the package when one crosses the line
from helpful to pedant. I find it useful to think of a (reviewer) time
budget, and maximizing the value I add to the package for the amount of
time I invest.
>
> the code given below is repeated many many times with slight variations
> for selected KEGG_path. i would ask the author
> to implement DRY (don't repeat yourself), find a functional approach to
> the repeated element, and
> simplify. i can imagine that damping enthusiasm of the contributor but
> I think it is worth holding back
> until such a thing is accomplished. too meddlesome?
>
> if (KEGG_path=="Carb_met") {
> pathways.list <- keggList("pathway", "hsa")## returns the list of human
> pathways
> mer<-select_path_carb(Carbohydrate)
> common<-intersect(pathways.list,mer)
> lo<-list()
> for (i in 1:length(pathways.list)){
> if (length(intersect(pathways.list[[i]],common)!=0)){
> lo[[i]]<-pathways.list[[i]]
> names(lo)[[i]]<-names(pathways.list)[[i]]
> }
> }
> pathways.list<-lo[lapply(lo,length)!=0]
> pathway.codes <- sub("path:", "", names(pathways.list))
> a<-do.call("rbind", pathways.list)
> }
>
>
> Another issue the author seems to raise is the need for "open cytoscape
> session" for meaningful tests.
> I think this will never happen on the build system. So the author has
> to emulate the cytoscape behavior
> perhaps with some static objects that cytoscape might return on interaction.
Yes, in earlier times the build system would be adjusted in various ways
to accommodate package needs, but this is not realistic.
Especially for tests, one wants to emphasize functionality of one's own
code and hence replace reliance on a service with proxy objects.
Martin
>
>
>
> On Fri, Sep 30, 2016 at 5:22 PM, Martin Morgan
> <martin.morgan at roswellpark.org <mailto:martin.morgan at roswellpark.org>>
> wrote:
>
> On 09/30/2016 12:05 PM, Michael Lawrence wrote:
>
> I have been chipping in here and there, but I'm not sure if I'm
> causing more confusion than help. I just select packages that
> interest
> me and contribute bite size pieces, but I would be happy to assume a
> more formal role. It takes very little of my time.
>
> One suggestion is that reviews should be incremental, i.e., provide
> obvious feedback first, let the package authors respond, then dig
> deeper for more detailed comments.
>
>
> Michael, I've found your comments / reviews really helpful. They are
> generally on-target and constructive, and help spur the authors to
> think about what they are doing. They do in some ways 'confuse' the
> issue (how is the package author to know that the 'assignee' is the
> official reviewer? are the people commenting on the package
> providing a consistent message?), but definitely a plus. I encourage
> others to participate in the public review process in this way, too.
>
> It would be great to hear from people who would like to take on the
> role of assigned reviewer. I'd suggest you identify package(s) that
> would be interesting to you, and send email to me. I'll arrange for
> you to become the assigned reviewer, and provide basic instructions
> and guidelines. In the future, you will be able to take on a package
> more directly.
>
> From the formal review process, incremental additional comments are
> not a good way to go, because the package author can feel like they
> are on a never-ending treadmill of requirements. Instead I think the
> formal review should more or less set out a contract -- do these
> things and your package will meet reviewer requirements for
> inclusion in Bioconductor. This makes it harder / more time
> consuming to review the package, because there is only the one
> chance to identify the most important points. Also, one needs to
> follow through and confirm that yes, the appropriate changes were made.
>
> It will be interesting to adjust the public review process,
> especially after the next release. For instance, one approach I like
> is to provide a review with editable check boxes, with the package
> author editing the review with information about commits, etc, that
> satisfy the concerns being expressed. I don't really like the canned
> check-list in the joss link Tim mentions; they point to important
> areas but seem too generic to lead to constructive changes to
> individual packages.
>
> Martin
>
>
> Michael
>
> On Fri, Sep 30, 2016 at 8:49 AM, Kasper Daniel Hansen
> <kasperdanielhansen at gmail.com
> <mailto:kasperdanielhansen at gmail.com>> wrote:
>
> One of the advantages of the Github review process is open
> review. We have
> some amount of packages submitted in the last moment (and I
> am guilty of
> one of these). Is help with the reviews needed? If so, how
> do we know
> where to concentrate?
>
> Kasper
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> Bioc-devel at r-project.org <mailto:Bioc-devel at r-project.org>
> mailing list
> https://stat.ethz.ch/mailman/listinfo/bioc-devel
> <https://stat.ethz.ch/mailman/listinfo/bioc-devel>
>
>
> _______________________________________________
> Bioc-devel at r-project.org <mailto:Bioc-devel at r-project.org>
> mailing list
> https://stat.ethz.ch/mailman/listinfo/bioc-devel
> <https://stat.ethz.ch/mailman/listinfo/bioc-devel>
>
>
>
> This email message may contain legally privileged and/or...{{dropped:2}}
>
>
> _______________________________________________
> Bioc-devel at r-project.org <mailto:Bioc-devel at r-project.org> mailing list
> https://stat.ethz.ch/mailman/listinfo/bioc-devel
> <https://stat.ethz.ch/mailman/listinfo/bioc-devel>
>
>
>
>
> --
>
> Vincent J. Carey, Ph.D.
> Professor of Medicine
> Channing Division of Network Medicine
> Brigham and Women's Hospital
> Harvard Medical School
> 181 Longwood Avenue
> Boston, MA 02115 USA
This email message may contain legally privileged and/or...{{dropped:2}}
More information about the Bioc-devel
mailing list