Without binmode strings with incompatible encoding can't be written in
the file. This is very common in applications that log user provided
parameters.
We need to allow changing the binnary mode because right now it is impossible to use
the built-in log rotation feature when you provide a File object to the
LogDevice, and if you provide a filename you can't have binmode.
9114b3ac7e
Thanks for the patch gareth (Gareth Adams). [Bug #15933]
-------
Combines two small, but very related changes
1: Treat HTTPS the same as HTTP
Previously, OpenURI followed guidance in RFC2616/3.7.1:
> When no explicit charset parameter is provided by the sender, media
> subtypes of the "text" type are defined to have a default charset
> value of "ISO-8859-1" when received via HTTP.
However this RFC was written before TLS was established and OpenURI was
never updated to treat HTTPS traffic the same way. So, HTTPS documents
received a different default to HTTP documents.
This commit removes the scheme check so that all text/* documents
processed by OpenURI are treated the same way.
In theory this processing gets applied to FTP URIs too, but there's no
mechanism in OpenURI for FTP documents to have Content-Type metadata
appended to them, so this ends up being a no-op.
2: Change default charset for text/* to UTF-8
Replaces the default ISO-8859-1 charset previously defined in RFC2616 (now
obsoleted) with a UTF-8 charset as defined in RFC6838.
Fixes: https://bugs.ruby-lang.org/issues/15933
Add encoding conversion (transcoding) from UTF-8 to CESU-8
and back. CESU-8 is an encoding similar to UTF-8, but encodes
codepoints above U+FFFF as two surrogates, these surrogates
again being encoded as if they were UTF-8 codepoints. This
preserves the same binary sorting order as in UTF-16. It is
also somewhat similar (although not exactly identical) to an
encoding used internally by Java.
This completes issue #15995.
enc/trans/cesu_8.trans: Add encoding conversion from/to CESU-8
test/ruby/test_transcode.rb: Add tests for above
WEBrick::HTTPProxyServer implementes HTTP proxy using
WEBrick and Net::HTTP.
WEBrick accepts HTTP/1.0 clients and
Net::HTTP uses always HTTP/1.1.
However HTTP/1.1 supports chunked transfer coding HTTP/1.0 doesn't.
Chunked transfer coding doesn't require that
content-length before the content is sent.
But non-chunked transfer coding require content-length before
the content is sent.
So, when HTTP/1.0 clients connects WEBrick::HTTPProxyServer and
origin server returns chunked response,
WEBrick::HTTPProxyServer needs to store whole content to
know the length of it.
This patch do it using tempfile.
As a comment token includes the newline, so delayed newline token
just follows it should not be dispatched. [Bug #11485]
Co-Authored-By: Jeremy Evans <code@jeremyevans.net>
It has caused CI failures.
* d0cd0866d8
Disable GC during rb_objspace_reachable_object_p
* 89cef1c56b
Version guard for [Feature #15974]
* 796eeb6339.
Fix up [Feature #15974]
* 928260c2a6.
Warn in verbose mode on defining a finalizer that captures the object
