hno [Thu, 29 May 2003 04:36:46 +0000 (04:36 +0000)]
Unified the bootstrap.sh scripts to make sure there is no mismatches
in autotool versions between the main sources and libTrie. As it was
the version used in libTrie was very old and failed to find compatible
versions if not the system default..
robertc [Wed, 21 May 2003 08:58:10 +0000 (08:58 +0000)]
Summary: Fix bad commit, and update test suite.
Keywords:
Astyling Makefile.am and cf.data.pre is bad... updated the reverted files to the intended content.
http_range_test's eventAdd dummy function needed updating to the new
signature.
cache_cf's parseBytesLine had a type in strcmp usage, preventing parsing.
robertc [Mon, 19 May 2003 15:11:30 +0000 (15:11 +0000)]
Summary: Really fixup HEAD builds.
Keywords:
DelayVector has incorrect case on update registration calls.
ACLChecklist inline methods were incorrect (or in the wrong file):
.cci files are for conditionally inline methods.
inline qualifier should not be used on conditionally inlined code.
as member functions matchAclList[Slow|Fast] need to be declared thusly.
The ACLChecklist header had not been updated to include the .cci file when
inlining was enabled.
robertc [Sun, 18 May 2003 06:03:55 +0000 (06:03 +0000)]
Summary: Various bugfixes.
Keywords:
* ACLChecklist::checkCallback: used deleteSelf() rather than delete this;
* ClientRequestContext::~ClientRequestContext: ditto.
* clientAccessCheckDone: Prevent race conditions leading to double deletes of acl_checklist.
* comm_accept_check_event: Add comment for clarity.
* StoreEntry::storeClientType: Add comments for clarity.
* StoreEntry::complete(): Ditto.
* Move .cci files to noinst_headers section.
This also includes a redefiniton of acl matching methods to make sure
there is no mismatches between fast/slow acl matches, and fixes
tcp_outgoing_* for the same..
For next version we should probably create a acl base class which
all the acl driven matches can use to limit the amount of duplicated
code in these acl driven types
allow/deny
tos
address
size
[direct/auto/peer/...] (to replace always_direct/never_direct/cache_peer_access)
All was Counter32 which is not correct for gauges.. Counter32 should only
be used for ever increasing counters. If unsure one of the Integer types
can be used, not Counter.
hno [Fri, 2 May 2003 01:46:51 +0000 (01:46 +0000)]
winbind_group 1.1 by Guido:
The changes are:
- Added an option for case insensitive group name comparation (Bugzilla #574)
- Fixed a segfault (Bugzilla #574)
- Updated the documentation according to FAQ on squid-users
This is the 9th of some splitted native Windows patches grouped by
functionality.
Native Windows port enhancements:
- changed opt_no_daemon to global
- now Cygwin can support daemon and Windows service mode in the same binary
- added Windows platform info in cache_manager output
- missing Makefile.am entry for squid_windows.h
- added more Windows service support definitions
- added cache.log info when unlinkd is disabled
- storeFsDone() is always needed for some fs storage: Coss, Awin32
Configure.in now only sets USE_EPOLL to 1 when EPOLL is the chosen comms interface. This means to build with epoll, you need --disable-poll --enable-epoll
Altered the use of USE_EPOLL from #ifdef to #if.
- new service support core functions
- changed platform identification from "Windows .NET" to "Windows Server 2003"
- include/squid_windows.h: New file, needed to prevent naming conflicts
between Windows includes and Squid
Create a blank squid-3.0 release notes, ready for fleshing out.
Change default created release notes to the 3.0 set.
Fill out the release notes from the CVS logs (50000 lines of review later!)
Still needs a final review and the remaining squid.conf changes copied/moved from the key changes section to the squid.conf section.
dhparams=/path/to/file.pem https_port option to specify DH parameters
for forward-secrecy in encryption (practically denies decryption even
if the private key is known from what I understand).
* Fix debug statements in client_side that where moved out of clientReadRequest to refer to their new function name.
* Refactor clientProcessBody into somewhat smaller chunks.
* Fix hung connections where more data was not retrieved in large POSTS.
Summary: Fix race when deleting clientReplyContext's.
Keywords:
Fix race when deleting clientReplyContext's. clientReplyContext::operator delete will cancel store reads and may trigger callbacks. However the object is no longer ready to handle those...