DeaR developers,
I was recently moving with my laptop between two environments with and without a
HTTP proxy server. As the internal proxy configuration is read only once from
the environment variables http_proxy/HTTP_PROXY at the first call of
download.file(), the proxy configurations couldn't be adjusted at a later
stage. (See also the comment in ?download.file). This caused some frustration as
I wasn't able to use http functionality in a running R session after moving
to an environment with a different/no proxy server.
After some digging I found that this behaviour was caused by RxmlNanoHTTPInit()
in src/modules/internet/nanohttp.c, which after its first call refuses to
reinitialise the proxy configuration. It does so by setting a variable
'initialized' to 1, and checking whether the string 'proxy' is
non-NULL.
I have made a small modification to have RxmlNanoHTTPInit() ignore these
variables and to read in the proxy configuration at each call.
$ svn diff
Index: src/modules/internet/nanohttp.c
==================================================================---
src/modules/internet/nanohttp.c (revision 62488)
+++ src/modules/internet/nanohttp.c (working copy)
@@ -255,15 +255,12 @@
WSADATA wsaData;
#endif
- if (initialized)
- return;
-
#ifdef _WINSOCKAPI_
if (WSAStartup(MAKEWORD(1, 1), &wsaData) != 0)
return;
#endif
- if (proxy == NULL) {
+ proxy = NULL;
proxyPort = 80;
env = getenv("no_proxy");
if (env && ((env[0] == '*') && (env[1] == 0)))
@@ -287,7 +284,7 @@
#endif
proxyUser = xmlMemStrdup(env);
}
- }
+
done:
initialized = 1;
}
This works fine for me (on a Mac) and I can simply use Sys.setenv(http_proxy=)
to adjust the proxy server at run time.
I think this doesn't really cause overhead as the number of calls do
download.file() is typically small and parsing the environment variables is
simple. On the other hand, if a user alters the proxy environment variables,
he/she probably does so for a reason. Googling around I found a few posts of R
users demanding this.
A cleaner solution would of course be to write a function that resets the
internal proxy configuration, but this would be a bit of an effort to make it
user visible.
What are your thoughts?
Best wishes,
Moritz
--
The Wellcome Trust Sanger Institute is operated by Genome Research
Limited, a charity registered in England with number 1021457 and a
company registered in England with number 2742969, whose registered
office is 215 Euston Road, London, NW1 2BE.
