When subsetting by column the eSet class, I came across this difficult to diagnose error: 

Error in `sampleNames<-`(`*tmp*`, value = c("910910", "1287381", "906862",  : 
  'value' length (965) must equal sample number in AssayData (965)

I discovered that it was due to the fact that I had previously set the assayData of an existing ExpressionSet object using `exprs<-`, while the object also had the se.exprs matrix filled. The exprs setting function let me mismatch the dimensions between exprs and se.exprs, leading to errors when later trying to subset the data. It seems like the exprs function should check for a mismatch in dimensions between exprs and se.exprs, and issue a warning when the matrix dimensions mismatch.

Thanks Biobase 2.33.1 (in devel; will be ported to release after seeing consequences in the builds) insists that the dimensions and dimnames of the incoming object match or can be made to match (e.g., NULL or sampled) those of the current object.

This has impact on my code (in minfi).  Specifically, in my class constructor, I create an object via "new()" and then subsequently modify the assayData elements.  This is done based on the profiling reported here: http://www.hansenlab.org/rstats/2014/01/30/lineprof

I can fix this relatively easily (at the cost of a performance overhead).  But to me, this is a case of whether we should allow creation of invalid objects, which can often be useful.  I think the proposed fix is too stringent, or I would like some way of disabling it for internal purpose (say force=FALSE default parameter to the replacement function).  But perhaps I am the only one affected and perhaps I am just too focused on my current issue.

Best,
Kasper