Hi Robert, Martin,
Yes using requireNamespace() internally is much cleaner than using
require(). Sorry for that. Just made the change in S4Vectors 0.2.6.
FYI the need to load IRanges namespace in a couple of places
inside S4Vectors is temporary and will go away soon.
Cheers,
H.
On 10/09/2014 09:33 AM, Martin Morgan wrote:
On 10/09/2014 08:00 AM, Robert Castelo wrote:
hi Martin,
thanks for your recommendations regarding the conditional loading of
packages. I
think however, that they are not related to the problem I'm referring.
Let me
put here a reproducible example which works with qpgraph version
0.99.7 that I
have just pushed to svn:
library(qtl)
library(qpgraph)
map <- sim.map(len=100, n.mar=10, anchor.tel=FALSE, eq.spacing=TRUE,
include.x=FALSE)
eqtlcross <- eQTLcross(map)
eqtlcross <- addGenes(eqtlcross, 5)
eqtlcross <- addeQTL(eqtlcross, "g1", location=map[[1]][1])
sim.eqtl <- reQTLcross(eqtlcross, rho=0.5, a=1)
cross <- sim.cross(map, sim.eqtl, n.ind=100)
gstarts <-runif(5, min=range(map[[1]])[1], max=range(map[[1]])[2])
annot <- data.frame(chr=rep(names(map)[1], 5),
start=gstarts, end=gstarts+1,
strand=rep("+", 5),
row.names=sim.eqtl$model$Y,
stringsAsFactors=FALSE)
## the following is the method that triggers the
## unexpected behavior. Its last but one instruction
## in line 208 of file qpgraph/R/eQTLnetworkEstimationParam-methods.R
## is the following:
##
## geneAnnotation <- geneAnnotation[genes]
##
## and should be using the method "[" imported from GenomicRanges
## however it starts loading a number of packages to do the job
param <- eQTLnetworkEstimationParam(cross, geneAnnotation=annot,
genome="simulatedGenome")
Loading required package: parallel
Attaching package: ?BiocGenerics?
The following objects are masked from ?package:parallel?:
[...etc...]
Hmm, I see that the parallel, BiocGenerics, and stats4 packages are
being attached. In a new R session I did this:
trace(loadNamespace, tracer=recover)
Tracing function "loadNamespace" in package "base"
[1] "loadNamespace"
param <- eQTLnetworkEstimationParam(cross, geneAnnotation=annot,
genome="simulatedGenome")
Loading required package: parallel
Tracing loadNamespace(package, c(which.lib.loc, lib.loc)) on entry
Enter a frame number, or 0 to exit
1: eQTLnetworkEstimationParam(cross, geneAnnotation = annot, genome =
"simulat
2: geneAnnotation[genes]
3: geneAnnotation[genes]
4: extractROWS(x, i)
5: extractROWS(x, i)
6: extractROWS(seqnames(x), i)
7: extractROWS(seqnames(x), i)
8: suppressWarnings(require(IRanges, quietly = TRUE))
9: withCallingHandlers(expr, warning = function(w)
invokeRestart("muffleWarnin
10: require(IRanges, quietly = TRUE)
...
and it's likely from
./S4Vectors/R/Rle-class.R: if (!suppressWarnings(require(IRanges,
quietly=TRUE)))
./S4Vectors/R/Rle-class.R: if (!suppressWarnings(require(IRanges,
quietly=TRUE)))
could be addressed with suppressPackageStartupMessages()
(less-preferred) or requireNamespace("IRanges") followed by
IRanges::...; let's see if we can get that fixed...
Martin
sessionInfo()
R version 3.1.0 (2014-04-10)
Platform: x86_64-unknown-linux-gnu (64-bit)
locale:
[1] LC_CTYPE=en_US.UTF8 LC_NUMERIC=C LC_TIME=en_US.UTF8
LC_COLLATE=en_US.UTF8
[5] LC_MONETARY=en_US.UTF8 LC_MESSAGES=en_US.UTF8
LC_PAPER=en_US.UTF8
LC_NAME=C
[9] LC_ADDRESS=C LC_TELEPHONE=C LC_MEASUREMENT=en_US.UTF8
LC_IDENTIFICATION=C
attached base packages:
[1] stats4 parallel stats graphics grDevices utils datasets
methods base
other attached packages:
[1] IRanges_1.99.28 S4Vectors_0.2.4 BiocGenerics_0.11.5
qpgraph_1.99.7 qtl_1.33-7
[6] vimcom_0.9-93 setwidth_1.0-3 colorout_1.0-2
loaded via a namespace (and not attached):
[1] annotate_1.43.5 AnnotationDbi_1.27.16 base64enc_0.1-2
BatchJobs_1.4
[5] BBmisc_1.7 Biobase_2.25.0 BiocParallel_0.99.22
biomaRt_2.21.1
[9] Biostrings_2.33.14 bitops_1.0-6 brew_1.0-6
checkmate_1.4
[13] codetools_0.2-9 DBI_0.3.1 digest_0.6.4
fail_1.2
[17] foreach_1.4.2 futile.logger_1.3.7 futile.options_1.0.0
GenomeInfoDb_1.1.23
[21] GenomicAlignments_1.1.30 GenomicFeatures_1.17.19
GenomicRanges_1.17.42
graph_1.43.0
[25] grid_3.1.0 iterators_1.0.7 lambda.r_1.1.6
lattice_0.20-29
[29] Matrix_1.1-4 mvtnorm_1.0-0 RCurl_1.95-4.3
Rgraphviz_2.9.1
[33] Rsamtools_1.17.34 RSQLite_0.11.4 rtracklayer_1.25.17
sendmailR_1.2-1
[37] stringr_0.6.2 tools_3.1.0 XML_3.98-1.1
xtable_1.7-4
[41] XVector_0.5.8 zlibbioc_1.11.1
cheers,
robert.
On 10/07/2014 05:54 PM, Martin Morgan wrote:
On 10/07/2014 08:15 AM, Robert Castelo wrote:
hi, it happens only with "[", that's why i'm puzzled.
it behaves as if you load a GRanges object 'x' and try to subset it
x[1]
without loading 'GenomicRanges' first.
Is there a reproducible example? I see in your code there are several
places where you require() or library() various packages. I think
one of
these Depends: on GenomicRanges, and the messages you see are the
effect
of moving GenomicRanges from 'loaded' to 'attached'. You can see the
effect with
library(qpgraph)
sessionInfo() ## GenomicRanges loaded but not attached
library(GenomicRanges) ## information about the package being attached
Probably in your code you do not actually want to require() ad hoc
packages and influence the user search path (and implicitly rely on
search path order for correct functionality), but rather to
requireNamespace("foo"); foo::fun(...) (or possibly loadNamespace()).
Complicated!
Martin
robert.
On 10/07/2014 05:05 PM, Michael Lawrence wrote:
Does that happen with the other methods or just "["? As a last
resort,
you could just drop the import (because "[" is a primitive, it should
just work).
On Tue, Oct 7, 2014 at 3:08 AM, Robert Castelo
<robert.castelo at upf.edu
<mailto:robert.castelo at upf.edu>> wrote:
hi Martin,
On 10/06/2014 07:24 PM, Martin Morgan wrote:
[...]
There are two 'as.vector' generics, one defined in Matrix and one in
BiocGenerics (and made available via IRanges). These generics have
different methods
showMethods(Matrix::as.vector)
Function: as.vector (package base)
x="abIndex", mode="ANY"
x="abIndex", mode="character"
x="ANY", mode="ANY"
x="dgCMatrix", mode="missing"
x="dgeMatrix", mode="missing"
x="diagonalMatrix", mode="missing"
x="dsCMatrix", mode="missing"
x="ldenseMatrix", mode="missing"
x="Matrix", mode="missing"
x="ndenseMatrix", mode="missing"
x="sparseVector", mode="character"
x="sparseVector", mode="missing"
showMethods(BiocGenerics::as.__vector)
Function: as.vector (package BiocGenerics)
x="ANY"
x="AtomicList"
x="Rle"
x="XDouble"
x="XInteger"
x="XRaw"
x="XString"
x="XStringSet"
so it's important that your code clearly distinguish between
generics.
One possibility is to remove importMethodsFrom(IRanges,
as.vector) from
the NAMESPACE, and explicitly use IRanges::as.vector(...) in
your code.
ok, i've done this as it is the easiest at the moment to meet the
release schedule. i guess that in the future i should try to avoid
using the '::' operator by importing exclusively what is needed from
each package.
codetoolsBioC::__writeNamespaceImports("__qpgraph") might
provide you with
some guidance (it's not 100% reliable; available via svn at
https://hedgehog.fhcrc.org/__bioconductor/trunk/madman/__Rpacks/codetoolsBioC
<https://hedgehog.fhcrc.org/bioconductor/trunk/madman/Rpacks/codetoolsBioC>)
about what functionality is being imported.
thanks for the heads up about codetoolsBioC, i've tried it out and
seen that some of the suggested imports are not necessary but some
others i was really missing them (which makes me wonder how was it
possible that he package did not break at those points).
one further question related to NAMESPACE. i subset GRanges objects
in the package via the '[' operator, i've included this into the
NAMESPACE file as:
importMethodsFrom(__GenomicRanges,
c, cbind, rbind,
"mcols<-", start, end, strand, sort,
"[", "[<-", "[[", "[[<-", "$", "$<-")
however, when the package reaches a subset operation x[i] with x
being a GRanges object, an entire package loading sequence starts:
Loading required package: GenomicRanges
Loading required package: BiocGenerics
Loading required package: parallel
Attaching package: ?BiocGenerics?
[... etc ...]
which may look a bit odd to the user. for every other imported
method the package uses them silently without loading the
corresponding package, am i importing '[' for GRanges objects from
the wrong package? is there a way to import '[' so that my package
can use it without triggering that package loading sequence?
thanks again!
robert.