[Bioc-devel] Unable to reproduce error!

[Martin Morgan]

On 10/20/2017 03:43 AM, Paolo Martini wrote:
> Dear all,
> 
> I am the clipper package maintainer.
> According to the "Build/check report" for BioC 3.6,
> the clipper package has an error while compiling the vignette.
> 
> I try to reproduce the error in my computer using the devel version but the
> R CMD check and installation goes smoothly.

I verified that I'm using devel and my packages are current

$ R
 > R.version.string           # 3.4.2
 > BiocInstaller::isDevel()   # TRUE
 > BiocInstaller::biocValid() # TRUE

I checked out a fresh copy of clipper, into a temporary directory

$ git clone https://git.bioconductor.org/packages/clipper

then installed it

$ R CMD INSTALL clipper

I then extracted the R source code of the vignette

$ cd clipper/vignettes
$ R CMD Stangle clipper.Rnw  # could have used knitr::purl() for Rmd

and tried to run the vignette code

$ R -f clipper.R

It fails as on the build system

 > pathwayAnalysis <- pathQ(exp, classes, graph, nperm=100, alphaV=0.05, 
b=100)
Error in getExpression(exprs(expr), classes) : Gene names not specified.
Calls: pathQ ... getExpression -> getExpression -> getExpression -> 
getExpression
Execution halted

OK, reproducible without any additional stuff like devtools or RStudio 
getting in the way. Time to debug. I started R in my favorite IDE, and 
tried to source the vignette

$ source("clipper.R", echo=TRUE, max=Inf)

it fails in the same way. I looked for the error, and tried to reproduce 
it in a simpler fashion, guessing a bit at what the failing code was 
trying to do ('expr' in the traceback() being the function argument 'exp')

 > pathwayAnalysis <- pathQ(exp, classes, graph, nperm=100, alphaV=0.05, 
b=100)
Error in getExpression(exprs(expr), classes) : Gene names not specified.
 > traceback()
10: stop("Gene names not specified.")
9: getExpression(exprs(expr), classes)
8: getExpression(exprs(expr), classes)
7: getExpression(expr, classes)
6: getExpression(expr, classes)
5: pathQ(exp, classes, graph, nperm = 100, alphaV = 0.05, b = 100) at 
clipper.R#95
4: eval(ei, envir)
3: eval(ei, envir)
2: withVisible(eval(ei, envir))
1: source("clipper.R", echo = TRUE, max = Inf)
 > clipper:::getExpression(exprs(exp), classes)
Error in clipper:::getExpression(exprs(exp), classes) :
   Gene names not specified.

I found that clipper:::getExpression is and S4 generic, with a likely 
method, so set the debugger and tried again

 > clipper:::getExpression
standardGeneric for "getExpression" defined from package "clipper"

function (expr, classes)
standardGeneric("getExpression")
<environment: 0x11061920>
Methods may be defined for arguments: expr, classes
Use  showMethods("getExpression")  for currently available ones.
 > showMethods(clipper:::getExpression)
Function: getExpression (package clipper)
expr="ExpressionSet", classes="numeric"
expr="matrix", classes="numeric"

 > debug(clipper:::getExpression, signature=c("matrix", "numeric"))
 > clipper:::getExpression(exprs(exp), classes)
Tracing clipper:::getExpression(exprs(exp), classes) step 2
Called from: eval(expr, p)
Browse[1]> n
debug: if (is.null(rownames(expr))) stop("Gene names not specified.")
Browse[2]> rownames(expr)
NULL
Browse[2]> Q
 >

and back at the top level

 > rownames(exprs(exp))
NULL

hmm, this looks weird

 > dim(exp)
Features  Samples
        0       79

In code chunk 3 you'd got the genes from graphite, and

 > head(genes)
[1] "ENTREZID:10000" "ENTREZID:1019"  "ENTREZID:1021"
[4] "ENTREZID:1026"  "ENTREZID:1029"  "ENTREZID:1147"

in code chunk 10 you'd calculated the intersection of these genes with 
the row.names of the sample expression set

 > intersect(genes, row.names(exprs(all)))
character(0)

and this is because the rownames are in a different format

 > head(row.names(exprs(all)))
[1] "5595" "7075" "1557" "643"  "643"  "1843"

I guess this is because graphite or one of it's dependencies has 
changed, accidentally or otherwise, the format of the identifiers 
returned in chunk 3. Likely the reason that you were not able to 
reproduce this is because you were using an out-of-date version of 
graphite (isDevel() and biocValid() would have diagnosed this). I don't 
know whether the resolution is a bug fix in graphite, or a fix to your 
own vignette code; it is always worth pushing the resolution as close to 
the source of the problem as possible, so if it's a change in graphite 
then engage the maintainer("graphite") in your investigation.

Martin

> 
> Anyone has suggestions on how to debug this?
> 
> 


This email message may contain legally privileged and/or...{{dropped:2}}