[
https://issues.apache.org/jira/browse/THRIFT-96?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12637650#action_12637650
]
Mark Slee commented on THRIFT-96:
---------------------------------
I believe the reason for doing read() instead of peek() with TBufferedTransport
was a mini-optimization. Since you're buffering the reads anyway, you might as
well actually pull the data into your buffer if you're going to check for its
existence. This way, you don't make a peek() call and a read() call, which is
beneficial in the default case of using a buffered socket. This means 1 syscall
to read() instead of 2 syscalls.
I'd be fine with switching this to use peek() on the underlying transport,
since in some cases read() may have side effects that aren't wanted on the
peek() operation (though I would expect this to be quite rare). But I'd also be
fine with making the interface contract specify that calls to read() must not
have any side effects that a peek() call would not also want.
> TSocket.peek fails on FreeBSD
> -----------------------------
>
> Key: THRIFT-96
> URL: https://issues.apache.org/jira/browse/THRIFT-96
> Project: Thrift
> Issue Type: Bug
> Components: Library (C++)
> Environment: FreeBSD
> Reporter: Alexander Shigin
> Attachments: thrift-freebsd-read.patch, thrift-peek-fix.patch,
> thrift-peek-fix.patch
>
>
> POSIX says what recv(2) should returns 0 if peer has performed a shutdown.
> This feature uses in TBufferedTransport
> {code}
> bool peek() {
> if (rBase_ == rBound_) {
> setReadBuffer(rBuf_.get(), transport_->read(rBuf_.get(), rBufSize_));
> }
> return (rBound_ > rBase_);
> }
> {code}
> The decision works fine on linux, but fails on freebsd. In freebsd, recv
> returns -1 and errno==ECONNRESET.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
