Leo Famulari <l...@famulari.name> writes: > On Wed, Sep 21, 2016 at 09:24:10PM +0200, Ricardo Wurmus wrote: >> Leo Famulari <l...@famulari.name> writes: >> >> > On Tue, Sep 13, 2016 at 11:53:33PM +0200, Roel Janssen wrote: >> >> This patch was essential to me being able to interact with HTTPS urls in >> >> R. As far as I understand, by default, R only looks for CURL_CA_BUNDLE >> >> on Windows, but with this patch it looks for CURL_CA_BUNDLE on GNU/Linux >> >> as well. Is this correct? >> >> >> >> I can confirm it works for me, so I'd like to see this patch pushed. >> > >> > It's good to hear that it works, but I still think we should run it by >> > the upstream maintainers. We are activating C code that they >> > specifically decided not to use on GNU / Linux. Why did they do that? >> >> The comments in the code indicate that on Windows they try to load the >> certs bundle that comes with R for Windows, i.e. in the R HOME’s “etc” >> directory. There is no such file on GNU, so no special handling is >> needed. >> >> On GNU this is taken care of by libcurl. It comes with a default path >> to the certs bundle, which can be overridden with configure flags >> (“--with-ca-bundle” or “--with-ca-path”). In our Guix package we don’t >> do this (yet?), so by default SSL cert validation is broken. >> >> libcurl does not respect CURL_CA_BUNDLE; it assumes that the application >> will override the CA bundle path if it needs a special path, otherwise >> it assumes that the default path is fine (using Guix this is not the >> case). >> >> The maintainers of the R curl package made the special case for Windows >> because it is not needed on GNU systems following the FHS. The best fix >> here would be to patch libcurl such that it checks the CURL_CA_BUNDLE >> environment variable invariably, just like the curl command line tool >> does. Until this is done I think we should path packages such as >> r-curl to make them usable. Once we have agreed on a fix to libcurl we >> can remove all patches to individual packages using libcurl. > > That makes sense. Thank you for taking the time to explain.
Sure! Pushed to master as 8f309571d3847d4bca331061e881fa01d9badb77. ~~ Ricardo
