httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "William A. Rowe, Jr." <wr...@lnd.com>
Subject RE: Four Apache-2.0 fixups
Date Tue, 11 Apr 2000 17:55:57 GMT


> -----Original Message-----
> From: Bill Stoddard [mailto:stoddard@raleigh.ibm.com]
> Sent: Tuesday, April 11, 2000 9:12 AM
> To: new-httpd@apache.org
> Subject: Re: Four Apache-2.0 fixups
>
>
> Can't we just call ap_oslevel to make a runtime check?  I'd
> really rather
> not use FindFirstFile when GetFileAttributesEx is available.  I've not
> tested it, but I am reasonably sure there will be a
> significant performance
> hit.

We can either delayload the whole project, binding all calls late to the dll
if they exist, thereby we die late upon encountering a call that doesn't
have the matching function name or ordinal.  We also slow beginning
execution, are are prone to 'hiccups' as untraveled code paths are first
executed, perhaps hours after the server is started.

Attached are the diffs to implement explicitly dynamic late-bound functions.
All this code was mangled from my run time dynamic Ansi/Unicode RCHARSTR
library coded in C++.  There are also some non-trivial issues with execution
order of static initialization, and forewarnings that
LoadLibrary/GetProcAddress cannot be executed within certain phases of the
load/fixup in the DLL code path.

It's tested under NT and behaves correctly, although I would like to know
what a thread exit code on termination of 0xC000013A really means.  I've
tried to eliminate the risks I mention above, but want to see it fault if
something fatal happens (should never fail if ap_get_oslevel tests are in
place - and I didn't care to slow it down further).  I have no time this
week to work on it further, but feel free to commit and start tweaking, and
add better failure handling (raise fatal exception with and intellegent
message, perhaps?).

Mime
View raw message