[Log4cplus-devel] flush in ConsoleAppender

[Tommi <t.m...@ep...>]

Hi,

I sent a few days ago a path, which flushes the ConsoleAppender after each=
=20
message. I need it, because I use std::ios::sync_with_stdio(false). The=20
functionality is now improved in the CVS-version. Like the file-appender=20
immediateFlush is now a option, which defaults to false.

But I think it does not make much sense to make it optional and worse -=20
defaulting to false. Standard is to output each single character written to=
=20
std::cout or std::cerr with a systemcall ::write. So even outputting "Hello=
"=20
results in 5 systemcalls! That is mandatory, because someone might mix=20
iostream-output with stdio-output (printf and friends). If you don't use=20
stdio, you can switch it off with std::ios::sync_with_stdio(false). That=20
improves vastly the performance of std::cout and std::cerr.

So in log4cplus the effect is in default configuration (immediateFlush to=20
false):
without std::ios::sync_with_stdio(false): each single character triggers a=
=20
system-call (very bad, but standard).
with std::ios::sync_with_stdio(false): my logmessages are delayed until the=
=20
buffer fills. In my application (a web-application-server) I don't see any=
=20
messages until I hav logged enough. That is somekind irritating. I log to=20
stdout for good reason: I want to see, what happens. But I don't see it,=20
because output is buffered.

When immediateFlus defaults to true (or is not optional at all) the effect =
is:
without std::ios::sync_with_stdio(false): no change
with std::ios::sync_with_stdio(false): each logmessage is immediatly writte=
n=20
to stdout in one single systemcall.

I don't see any reason to switch immediateFlush off. It just delays=20
logmessages needlessly.


Tommi M=E4kitalo