R devel - Sep 2024 - Can gzfile be given the same method option as file

Recently my employer has introduced a security system which generates SSL
certificates on the fly to be able to see the content of https connections.  To
make this work they add a new root certificate to the windows certificate store.

In R this causes problems because the default library used to download data from
URLs doesn't look at this store, however the "wininet" download
method works so where this is used then things work (albeit with a warning about
future deprecation).

For functions like download.file this works great, but it fails when running
readRDS:

readRDS('https://seurat.nygenome.org/azimuth/references/homologs.rds')
Error in gzfile(file, "rb") : cannot open the connection
In addition: Warning message:
In gzfile(file, "rb") :
  cannot open compressed file
'https://seurat.nygenome.org/azimuth/references/homologs.rds', probable
reason 'Invalid argument'

After some debugging I see that the root cause is from the gzfile function.
>
gzfile('https://seurat.nygenome.org/azimuth/references/homologs.rds')
-> g
> open(g, open="r")
Error in open.connection(g, open = "r") : cannot open the connection
In addition: Warning message:
In open.connection(g, open = "r") :
  cannot open compressed file
'https://seurat.nygenome.org/azimuth/references/homologs.rds', probable
reason 'Invalid argument'

If this was not a compressed file then using file rather than gzfile we can make
this work by setting the url.method option:
> options("url.method"="wininet")
> file('https://seurat.nygenome.org/azimuth/references/homologs.rds')
-> g
> open(g, open="r")
Warning message:
In open.connection(g, open = "r") :
  the 'wininet' method of url() is deprecated for http:// and https://
URLs

So I get a warning, but it works.

I guess this boils down to two questions:


  1.  Is it possible to add the same "method" argument to gzfile that
file uses so that people in my situation have a work round?
  2.  Given the warnings we're getting when using wininet, are their plans
to make windows certficates be supported in another way?

Thanks

Simon.












	[[alternative HTML version deleted]]

? Thu, 12 Sep 2024 12:01:54 +0000
Simon Andrews via R-devel <r-devel at r-project.org> ?????:
>
readRDS('https://seurat.nygenome.org/azimuth/references/homologs.rds')
> Error in gzfile(file, "rb") : cannot open the connection
I don't think that gzfile works with URLs. gzcon(), on the other hand,
does work with url() connections, which accepts the 'method' argument
and the getOption('url.method') default.

h <- readRDS(url(
 'https://seurat.nygenome.org/azimuth/references/homologs.rds'
))

But that only works with gzip-compressed files. For example, CRAN's
PACKAGES.rds is xz-compressed, and I don't see a way to read it the
same way:

readBin(
 index <- file.path(
  contrib.url(getOption('repos')['CRAN']),
  'PACKAGES.rds'
 ), raw(), 5
) |> rawToChar()
# [1] "\xfd7zXZ" <-- note the "7zXZ" header
readRDS(url(index))
# Error in readRDS(url(index)) : unknown input format
>   2.  Given the warnings we're getting when using wininet, are their
> plans to make windows certficates be supported in another way?
What does libcurlVersion() return for you? In theory, it should be
possible to make libcurl use schannel and therefore the system
certificate store for TLS verification purposes.

-- 
Best regards,
Ivan