[Bioc-devel] Windows, normalizePath(), and non-ASCII characters

[Mike Smith]

In trying to diagnose this issue at
https://support.bioconductor.org/p/108548/ I've found some weird behaviour
with Windows, normalizePath(), and non-ASCII characters.  Essentially, if I
run normalizePath() recursively on a path that contains  'é' (I haven't
tried other characters) something 'changes' in the string, but I can't work
out what, and it breaks a subsequent .Call() which uses the path.

The example below tries to demonstrate this in a fairly concise manner. It
works fine if normalizePath() is run once, but fails after it's run a
second time on itself.

However, change "éxample" for "example" and both instances work. Similarly,
both run fine on my Linux machine with the non-ASCII character inplace.

I'd be grateful if anyone else with a Windows machine could verify this
behaviour, or to shed any light on what might be the difference between path1
and path2 below.

Thank,
Mike

------------------------------

## setup some HDF5 components required later
flags <- rhdf5:::h5checkConstants("H5F_ACC", h5default("H5F_ACC"))
fcpl <- rhdf5:::h5checktypeAndPLC(NULL, "H5P_FILE_CREATE", allowNULL = TRUE)
fapl <- rhdf5::H5Pcreate("H5P_FILE_ACCESS")

## create a folder with non-ASCII character
dir.create('éxample')
setwd("éxample")

## create two normalized paths recursively - these are 'identical'
path1 <- normalizePath('test.h5', mustWork = FALSE)
path2 <- normalizePath(path1, mustWork = FALSE)
identical(path1, path2)

## create an HDF5 file using path1 - this works
fid <- .Call("_H5Fcreate", path1, flags, fcpl at ID, fapl at ID,
             PACKAGE = "rhdf5")
.Call("_H5Fclose", fid, PACKAGE = "rhdf5")
file.remove(path1)

## create an HDF5 file using path2 - this fails
fid <- .Call("_H5Fcreate", path2, flags, fcpl at ID, fapl at ID,
             PACKAGE = "rhdf5")
if(exists('fid2')) {
  .Call("_H5Fclose", fid2, PACKAGE = "rhdf5")
  file.remove(path2)
}

## tidy up
rhdf5::h5closeAll()
setwd("../")

	[[alternative HTML version deleted]]