CHANGES

Changes with Apache 2.0a9

  *) Win32 now requires perl to complete the final install step for users
     to build + install on Win32.  Makefile.win now rewrites @@ServerRoot@
     and installs the conf, htdocs and htdocs/manual directories.
     [William Rowe]

  *) Make mod_include use a hash table to associate directive tags with
     functions.  This allows modules to implement their own SSI tags easily.
     The idea is simple enough, a module can insert it's own tag and function
     combination into a hash table provided by mod_include.  While mod_include
     parses an SSI file, when it encounters a tag in the file, it does a
     hash lookup to find the function that implements that tag, and passes
     all of the relevant data to the function.  That function is then
     responsible for processing the tag and handing the remaining data back
     to mod_include for further processing.
     [Paul J. Reder <rederpj@raleigh.ibm.com>]

  *) Get rid of ap_new_apr_connection().  ap_new_connection() now has 
     fewer parameters: the local and remote socket addresses were removed
     from the parameter list because all required information is available
     via the APR socket.  [Jeff Trawick]

  *) Distribution directory structure reorganized to reflect a
     normal source distribution with external install targets.
     [Roy Fielding]

  *) The MPMs that need multiple segments of shared memory now create
     two apr_shmem_t variables, one for each shared memory allocation.
     the problem is that we can't determine how much memory will be required
     for shared memory allocations once we try to allocate more than one
     variable.  The MM code automatically aligns the shared memory allocations,
     so we end up needing to pad the amount of shared memory we want based
     on how many variables will be allocated out of the shared memory segment.
     It is just easier to create a second apr_shmem_t variable, and two
     shmem memory blocks.
     [Ryan Bloom]

  *) Cleanup the export list a bit.  This creates a single unified list of
     functions exported by APR.  The export list is generated at configure
     time, and that list is then used to generate the exports.c file.
     Because of the way the export list is generated, we only export those
     functions that are valid on the platform we are building on.
     [Ryan Bloom]

  *) Enable logging the cookie with mod_log_config
     [Sander van Zoest <sander@covalent.net>]

  *) Fix a segfault in mod_info when it reaches the end of the configuration.
     [Jeff Trawick]

  *) Added lib/aputil/ as a placeholder for utility functions which are not
     specific to the Apache HTTP Server (but do not make sense with APR).
     The first utility is "apu_dbm": a set of functions to work with DBM
     files. This first version can be compiled for SDBM or GDBM databases.
     [Greg Stein]

  *) Complete re-write of mod_include.  This makes mod_include a filter that
     uses buckets directly.  This has now served the FAQ correctly.
     [Paul Reder <rederpj@raleigh.ibm.com>]

  *) Allow modules to specify the first filter in a sub_request when
     making the sub_request.  This keeps modules from having to change the
     output_filter immediately after creating the sub-request, and therefore
     skip the sub_req_output_filter.  [Ryan Bloom]

  *) Update ab to accept URLs with IPv6 literal address strings (in the
     format described in RFC 2732), and to build Host header fields in
     the same format.  This allows IPv6 literal address strings to be
     used with ab.  This support has been tested against Apache 1.3 with 
     the KAME patch, but Apache 2.0 does not yet work with this format
     of the Host header field.  [Jeff Trawick]

  *) Accomodate an out-of-space condition in the piped logs and the
     rotatelogs.c code, and no longer churn log processes for this
     condition.  [Victor J. Orlikowski]

  *) Add support for partial writes with apr_sendfile() to core_output_filter.
     [Greg Ames] 

Changes with Apache 2.0a8

  *) Add a directive to mod_mime so that filters can be associated with
     a given mime-type.
     [Ryan Bloom]

  *) Get multi-views working again.  We were setting the path_info
     field incorrectly if we couldn't find the specified file.
     [Ryan Bloom]

  *) Fix 304 processing.  The core should never try to send the headers
     down the filter stack.  Always, just setup the table in the request
     record, and let the header filter convert it to data that is ready
     for the network.
     [Ryan Bloom]

  *) More fixes for the proxy.  There are still bugs in the proxy code,
     but this has now proxied www.yahoo.com and www.ntrnet.net (my ISP)
     successfully.
     [Ryan Bloom]

  *) Fix params for apr_getaddrinfo() call in connect proxy handler.
     [Chuck Murcko]

  *) APR: Add new apr_getopt_long function to handle long options.
     [B. W. Fitzpatrick <fitz@red-bean.com>]

  *) APR: Change apr_connect() to take apr_sockaddr_t instead of hostname.
     Add generic apr_create_socket().  Add apr_getaddrinfo() for doing
     hostname resolution/address string parsing and building
     apr_sockaddr_t.  Add apr_get_sockaddr() for getting the address
     of one of the apr_sockaddr_t structures for a socket.  Change
     apr_bind() to take apr_sockaddr_t.  [David Reid and Jeff Trawick]

  *) Remove the BUFF from the HTTP proxy.  This is still a bit ugly, but
     I have proxied pages with it, cleanup will commence soon.
     [Ryan Bloom]

  *) Make the proxy work with filters.  This isn't perfect, because we
     aren't dealing with the headers properly.  [Ryan Bloom]

  *) Do not send a content-length iff the C-L is 0 and this is a head
     request.  [Ryan Bloom]

  *) Make cgi-bin work as a regular directory when using mod_vhost_alias
     with no VirtualScriptAlias directives. PR#6829 [Tony Finch]

  *) Remove BUFF from the PROXY connect handling. [Ryan Bloom]

  *) Get the default_handler to stop trying to deal with HEAD requests.
     The idea is to let the content-length filter compute the C-L before
     we try to send the data.  If we can get the C-L correctly, then we
     should send it in the HEAD response.
     [Ryan Bloom]
     
  *) The Header filter can now determine if a body should be sent based
     on r->header_only.  The general idea of this is that if we delay
     deciding to send the body, then we might be able to compute the
     content-length correctly, which will help caching proxies to cache
     our data better.  Any handler that doesn't want to try to compute
     the content-length can just send an EOS bucket without data and
     everything will just work.
     [Ryan Bloom]

  *) Add the referer to the error log if one is available.
     [Markus Gyger <mgyger@itr.ch>]

  *) Mod_info.c has now been ported to Apache 2.0.  As a part of this
     change, the root of the configuration tree has been exposed to modules
     as ap_conftree.
     [Ryan Morgan <rmorgan@covalent.net>]

  *) Get the core_output_filter to use the bucket interface directly.
     This keeps us from calling the content-length filter multiple times
     for a simple static request.
     [Ryan Bloom]

  *) We are sending the content-type correctly now.
     [Ryan Bloom and Will Rowe]

  *) APR on FreeBSD: Fix a bug in apr_sendfile() which caused us to report
     a bogus bytes-sent value when the only thing being sent was trailers
     and writev() returned an error (or EAGAIN).  [Jeff Trawick]

  *) Get SINGLE_LISTEN_UNSERIALIZED_ACCEPT working again.  This uses the
     hints file to determine which platforms define 
     SINGLE_LISTEN_UNSERIALIZED_ACCEPT.
     [Ryan Bloom]

  *) APR: add apr_get_home_directory()  [Jeff Trawick]

  *) Initial import of 1.3-current mod_proxy. [Chuck Murcko]

  *) Not all platforms have INADDR_NONE defined by default.  Apache
     used to make this check and define INADDR_NONE if appropriate,
     but APR needs the check too, and I suspect other applications will
     as well.  APR now defines APR_INADDR_NONE, which is always a valid
     value on all platforms.
     [Branko Èibej <brane@xbc.nu>]

  *) Destroy the pthread mutex in lock_intra_cleanup() for PR#6824.
     [Shuichi Kitaguchi <ki@hh.iij4u.or.jp>] 

  *) Relax the syntax checking of Host: headers in order to support
     iDNS. PR#6635 [Tony Finch]

  *) When reading from file buckets we convert to an MMAP if it makes
     sense.  This also simplifies the default handler because the
     default handler no longer needs to try to create MMAPs.
     [Ryan Bloom]

  *) BUFF has been removed from the main server.  The BUFF code will remain
     in the code until it has been purged from the proxy module as well.
     [Ryan Bloom]

  *) Byteranges have been completely re-written to be a filter.  This
     has been tested, and I believe it is working correctly, but it could
     doesn't work for the Adobe Acrobat plug-in.  The output almost matches
     the output from 1.3, the only difference being that 1.3 includes
     a content-length in the response, and this does not.
     [Ryan Bloom]

  *) APR read/write functions and bucket read functions now operate
     on unsigned integers, instead of signed ones.  It doesn't make
     any sense to use signed ints, because we return the error codes,
     so if we have an error we should report 0 bytes read or written.
     [Ryan Bloom]

  *) Always compute the content length, whether it is sent or not.
     The reason for this, is that it allows us to correctly report
     the bytes_sent when logging the request.  This also simplifies
     content-length filter a bit, and fixes the actual byte-reporing
     code in mod_log_config.c
     [Ryan Bloom]

  *) Remove AP_END_OF_BRIGADE definition.  This does not signify what
     it says, because it was only used by EOS and FLUSH buckets.  Since
     neither of those are required at the end of a brigade, this was
     really signifying FLUSH_THE_DATA, but that can be determined better
     by checking AP_BUCKET_IS_EOS() or AP_BUCKET_IS_FLUSH.  EOS and FLUSH
     buckets now return a length of 0, which is actually the amount of data
     read, so they make more sense.
     [Ryan Bloom]

  *) Allow the core_output_filter to save some data past the end of a
     request.  If we get an EOS bucket, we only send the data if it 
     makes sense to send it.  This allows us to pipeline request
     responses.  As a part of this, we also need to allocate mmap
     buckets out of the connection pool, not the request pool.  This
     allows the mmap to outlive the request.
     [Ryan Bloom]

  *) Make blocking and non-blocking bucket reads work correctly for
     sockets and pipes.  These are the only bucket types that should
     have non-blocking reads, because the other bucket types should
     ALWAYS be able to return something immediately.
     [Ryan Bloom]

  *) In the Apache/Win32 console window, accept Ctrl+C to stop the 
     server, but use Ctrl+Break to initiate a graceful restart 
     instead of duplicating behavior. [John Sterling]

  *) Patch mod_autoindex to set the Last-Modified header based on
     the directory's mtime, and add the ETag header.  [William Rowe]

  *) Merge the 1.3 patch to add support for logging query string in 
     such a way that "%m %U%q %H" is the same as "%r".
     [Bill Stoddard]

  *) Port three log methods from mod_log_config 1.3 to 2.0: 
     CLF compliant '-' byte count, method and protocol.
     [Bill Stoddard]

  *) Add a new LogFormat directive, %c, that will log connection
     status at the end of the response as follows:
     'X' - connection aborted before the response completed.
     '+' - connection may be kept-alive by the server.
     '-' - connection will be closed by the server.
     [Bill Stoddard]

  *) Expand APR for WinNT to fully accept and return utf-8 encoded
     Unicode file names and paths for Win32, and tag the Content-Type 
     from mod_autoindex to reflect that charset if the the feature
     macro APR_HAS_UNICODE_FS is true.  [William Rowe]

  *) Compute the content length (and add appropriate header field) for
     the response when no content length is available and we can't use 
     chunked encoding.  [Jeff Trawick]

  *) Changed ap_discard_request_body() to use REQUEST_CHUNKED_DECHUNK,
     so that content input filters get dechunked data when using
     the default handler. Also removed REQUEST_CHUNKED_PASS.
     [Sascha Schumann]
     
  *) Add mod_ext_filter as an experimental module.  This module allows
     the administrator to use external programs as filters.  Currently,
     only filtering of output is supported.  [Jeff Trawick]

  *) Most Apache functions work on EBCDIC machines again, as protocol
     data is now translated (again).  [Jeff Trawick]

  *) Introduce ap_xlate_proto_{to|from}_ascii() to clean up some of
     the EBCDIC support.  They are noops on ASCII machines, so this
     type of translation doesn't have to be surrounded by #ifdef
     CHARSET_EBCDIC.  [Jeff Trawick]

  *) Fix mod_include.  tag commands work again, and the server will
     send the FAQ again.  This also allows mod_include to set aside
     buckets that include partial buckets.
     [Ryan Bloom and David Reid]

  *) Add suexec support back.  [Manoj Kasichainula]

  *) Lingering close now uses the socket directly instead of using
     BUFF.  This has been tested, but since all we can tell is that it
     doesn't fail, this needs to be really hacked on.
     [Ryan Bloom]

  *) Allow filters to modify headers and have those headers be sent to
     the client.  The idea is that we have an http_header filter that
     actually sends the headers to the network.  This removes the need
     for the BUFF to send headers.
     [Ryan Bloom]

  *) Charset translation: mod_charset_lite handles translation of
     request bodies.  Get rid of the xlate version of ap_md5_digest()
     since we don't compute digests of filtered (e.g., translated) 
     response bodies this way anymore.  (Note that we don't do it at
     all at the present; somebody needs to write a filter to do so.)
     [Jeff Trawick]

  *) Input filters and ap_get_brigade() now have a input mode parameter 
     (blocking, non-blocking, peek) instead of a length parameter.
     [hackathon]

  *) Update the mime.types file to the registered media types as
     of 2000-10-19. PR#6613 [Carsten Klapp <carsten.klapp@home.net>,
     Tony Finch]

  *) Namespace protect some macros declared in ap_config.h
     [Ryan Bloom]

  *) Support HTTP header line folding with input filtering.
     [Greg Ames]

  *) Mod_include works again.  This should still be re-written, but at
     least now we can serve an SHTML page again.
     [Ryan Bloom]

  *) Begin to remove BUFF from the core.  Currently, we keep a pointer
     to both the BUFF and the socket in the conn_rec.  Functions that
     want to use the BUFF can, functions that want to use the socket,
     can.  They point to the same place.
     [Ryan Bloom]

  *) apr_psprintf doesn't understand %lld as a format.  Make it %ld.
     [Tomas "Ögren" <stric@ing.umu.se>]

  *) APR pipes on Unix and Win32 are now cleaned up automatically when the 
     associated pool goes away.  (APR pipes on OS/2 were already had this
     logic.)  This resolvs a fatal file descriptor leak with CGIs.  
     [Jeff Trawick]

  *) The final line of the config file was not being read if there was
     no \n at the end of it.  This was caused by apr_fgets returning 
     APR_EOF even though we had read valid data.  This is solved by
     making cfg_getline check the buff that was returned from apr_fgets.
     If apr_fgets return APR_EOF, but there was data in the buf, then we
     return the buf, otherwise we return NULL.
     [Ryan Bloom]

  *) Piped logs work again in the 2.0 series.
     [Ryan Bloom]

  *) Restore functionality broken by the mod_rewrite security fix:
     rewrite map lookup keys and default values are now expanded
     so that the lookup can depend on the requested URI etc.
     PR #6671 [Tony Finch]

  *) Tighten up the syntax checking of Host: headers to fix a
     security bug in some mass virtual hosting configurations
     that can allow a remote attacker to retrieve some files
     on the system that should be inaccessible. [Tony Finch]

  *) Add a pool bucket type.  This bucket is used for data allocated out
     of a pool.  If the pool is cleaned before the bucket is destroyed, then
     the data is converted to a heap bucket, allowing it to survive the
     death of the pool.
     [Ryan Bloom]

  *) Add a flush bucket.  This allows modules to signal that the filters
     should all flush whatever data they currently have.  There is no way
     to actually force them to do this, so if a filter ignores this bucket,
     that's life, but at least we can try with this.
     [Ryan Bloom]

  *) Add an output filter for sub-requests.  This filter just strips the
     EOS bucket so that we don't confuse the main request's core output
     filter by sending multiple EOS buckets.  This change also makes sub
     requests start to send EOS buckets when they are finished.
     [Ryan Bloom]

  *) Make ap_bucket_(read|destroy|split|setaside) into macros.  Also
     makes ap_bucket_destroy a return void, which is okay because it
     used to always return APR_SUCCESS, and nobody ever checked its
     return value anyway.
     [Cliff Woolley <cliffwoolley@yahoo.com>]

  *) Remove the index into the bucket-type table from the buckets
     structure.  This has now been replaced with a pointer to the
     bucket_type.  Also add some macros to test the bucket-type.
     [Ryan Bloom]

  *) Renamed all MODULE_EXPORT symbols to AP_MODULE_DECLARE and all symbols
     for CORE_EXPORT to AP_CORE_DECLARE (namespace protecting the wrapper)
     and retitled API_EXPORT as AP_DECLARE and APR_EXPORT as APR_DECLARE.
     All _VAR_ flavors changes to _DATA to be absolutely clear.
     [William Rowe]

  *) Add support for /, //, //servername and //server/sharename 
     parsing of <Directory> blocks under Win32 and OS2.
     [Tim Costello, William Rowe, Brian Harvard]

  *) Remove the function pointers from the ap_bucket type.  They have been
     replaced with a global table.  Modules are allowed to register bucket
     types and use then use those buckets.
     [Ryan Bloom]

  *) mod_cgid: In the handler, shut down the Unix socket (only for write) 
     once we finish writing the request body to the cgi child process; 
     otherwise, the client doesn't hit EOF on stdin.  Small request bodies 
     worked without this change (for reasons I don't understand), but large 
     ones didn't.  [Jeff Trawick]

  *) Remove file bucket specific information from the ap_bucket type.
     This has been moved to a file_bucket specific type that hangs off
     the data pointer in the ap_bucket type.
     [Ryan Bloom]

  *) Input filtering now has a third argument.  This is the amount of data
     to read from lower filters.  This argument can be -1, 0, or a positive
     number.  -1 means give me all the data you have, I'll deal with it and
     let you know if I need more.  0 means give me one line and one line
     only.  A positive number means I want no more than this much data.

     Currently, only 0 and a positive number are implemented.  This allows
     us to remove the remaining field from the conn_rec structure, which
     has also been done.
     [Ryan Bloom] 
    
  *) Big cleanup of the input filtering.  The goal is that http_filter
     understands two conditions, headers and body.  It knows where it is
     based on c->remaining.  If c->remaining is 0, then we are in headers,
     and http_filter returns a line at a time.  If it is not 0, then we are
     in body, and http_filter returns raw data, but only up to c->remaining
     bytes.  It can return less, but never more.
     [Greg Ames, Ryan Bloom, Jeff Trawick]

  *) mod_cgi: Write all of the request body to the child, not just what
     the kernel would accept on the first write.  [Jeff Trawick]

  *) Back out the change that moved the brigade from the core_output_filters
     ctx to the conn_rec.  Since all requests over a given connection
     go through the same core_output_filter, the ctx pointer has the
     correct lifetime.
     [Ryan Bloom]

  *) Fix another bug in the send_the_file() read/write loop. A partial
     send by apr_send would cause unsent data in the read buffer to
     get clobbered. Complete making send_the_file handle partial
     writes to the network.
     [Bill Stoddard]

  *) Fix a couple of type fixes to allow compilation on AIX again
     [Victor J. Orlikowski <v.j.orlikowski@gte.net>]

  *) Fix bug in send_the_file() which causes offset to be ignored
     if there are no headers to send.
     [Bill Stoddard]

  *) Handle APR_ENOTIMPL returned from apr_sendfile in the core
     filter. Useful for supporting Windows 9* with a binary
     compiled on Windows NT.
     [Bill Stoddard]

Changes with Apache 2.0a7

  *) Reimplement core_output_filter to buffer/save bucket brigades
     across multiple calls to the core_filter. The brigade will be
     sent when either MIN_BYTES_TO_SEND or MAX_IOVEC_TO_WRITE
     thresholds are hit or the EOS bucket is received.
     [Bill Stoddard]

  *) Create experimental filter (buffer_filter) that coalesces bytes 
     into one large buffer before invoking the next filter in the
     chain. This filter is particularly useful with the current 
     implementation of mod_autoindex when it inserted above the
     chunk_filter. mod_autoindex generates a lot of brigades that
     containing buckets holding just a few bytes each. The
     buffer_filter coalesces these buckets into a single large bucket.
     [Bill Stoddard]

  *) Add apr_sendfile() support into the core_output_filter.
     [Bill Stoddard]

  *) Add apr_sendv() support into the core_output_filter.
     [Bill Stoddard]

  *) Fix mod_log_config so that it compiles cleanly with BUFFERED_LOGS
     [Mike Abbott <mja@sgi.com>]

  *) Remove ap_send_fb.  This is no longer used in Apache, and it doesn't
     make much sense, because Apache uses buckets instead of BUFFs now.
     [Ryan Bloom]

  *) send_the_file now falls back to a read/write loop on platforms that
     do not have sendfile.
     [Ryan Bloom and Brian Havard]

  *) Install apachectl correctly, and substitute the proper values so
     that it works again.  [Ryan Bloom]

  *) Better(??) handle platforms that lack sendfile().
     [Jim Jagielski]

  *) APR now has UUID generation/formatting/parsing support.
     [Greg Stein]

  *) Begin the http_filter.  This is an input filter that understands
     the absolute basic amount required to parse an HTTP Request.  The
     goal is to be able to split headers from request body before passing
     the data back to the other filters.
     [Ryan Bloom]

  *) Bring forward from 1.3.13 the config directory implementation
     [Jim Jagielski]

  *) install apxs if it is created
     [Ryan Bloom]

  *) Added APR_IS_STATUS_condition test macros to eliminate canonical error
     conversions.  [William Rowe]

  *) Now that we have ap_add_input_filter(), rename ap_add_filter() to 
     ap_add_output_filter().  [Jeff Trawick]

  *) Multiple build and configuration fixes
    Build process:

      -add datadir and localstatedir substitutions
      -fix layout name
      -fix logfilename misspelling
      -fix evaluation of installation dir variables and
      -replace $foobar by $(foobar) to be usefull in the makefile
    
    Cross compile:
    
      -add rules for cross-compiling in rules.mk. Okay, rule to check for
       $CC_FOR_BUILD is still missing
      -use CHECK_TOOL instead of CHECK_PROG for ranlib
      -add missing "AR=@AR@" to severaly Makefile.in's
      -cache result for "struct rlimit"
      -compile all helper programs with native and cross compiler
       and use the native version to generate header file
     ["Rüdiger" Kuhlmann <Tadu@gmx.de>]

  *) Prepare our autoconf setup for autoconf 2.14a and for cross-
     compiling.
     ["Rüdiger" Kuhlmann <Tadu@gmx.de>]

  *) Fix a bug where a client which only sends \n to delimit header
     lines (netcat) gets a strange looking HTTP_NOT_IMPLEMENTED 
     message.  Start working on ebcdic co-existance with input 
     filtering.
     [William Rowe, Greg Ames]

  *) If mod_so is enabled in the server always create libexec, even
     if there are no modules installed in this directory.  This is a
     requirement for APXS to work correctly.
     [Ryan Bloom]

  *) Connection oriented output filters are now stored in the 
     conn_rec instead of the request_rec.  This allows us to add the
     output filter in the pre-connection phase instead of the
     post_read_request phase, which keeps us from trying to write an
     error page before we have a filter to write to the network.
     [Ryan Bloom, Jeff Trawick, and Greg Ames]

  *) Cleaning up an mmap bucket no longer deletes the mmap.  An
     mmap can be used across multiple buckets (default_handler with
     byte ranges, mod_file_cache, mod_mmap_static), so cleanup of
     the mmap itself can't be associated with the bucket.
     [Jeff Trawick]

  *) Add .dll caching directive ISAPICacheFile to mod_isapi.
     [William Rowe]

  *) Radical surgery to improve mod_isapi support under Win32.
     Includes a number of newer ServerSupportFunction calls, support
     for ReadClient (in order to retrieve POSTs greater than 48KB),
     and general bug fixes to more reliably load ISAPI .dll's and
     prevent leaking handle resources.  Note: There are still 
     discrepancies between IIS's and Apache's ServerVariables, and
     async calls are still not supported.  Additional warnings are
     logged to facilitate debugging of unsupported ISAPI calls.
     [William Rowe]

  *) Add input filtering to Apache.  The basic idea for the input
     filters is the same as the ideas for output filters.  The biggest
     difference is that instead of calling ap_pass_brigade, ap_get_brigade
     should be called, and the order of execution for the filter itself is
     different.  When writing an output filter, a brigade is passed in,
     and filters operate directly on that brigade, when done, they call
     ap_pass_brigade.  Input filters are the exact opposite.  Because input
     is not a push operation, filters first call ap_get_brigade.  When this
     function returns, the input filter will be left with a valid brigade.
     The input filter should then operate on the brigade, and return.
     [Ryan Bloom]

  *) Fix building on BSD/OS using its native make. The build system
     falls back to the BSD .include directive on that host platform.
     [Sascha Schumann]

  *) Expand dbmmanage to allow -d -m -s -p options for Crypt, MD5,
     SHA1 and plaintext password encodings.  Make feature tests a
     bit more flexible.  [William Rowe]

  *) Charset translation: mod_charset_lite handles output content 
     translation in a filter.  mod_charset_lite no longer ignores 
     subrequests.  A bunch of cruft related to BUFF's support for
     translating request and response bodies was removed.  
     [Jeff Trawick]

  *) Move the addition of the CORE filter to the post_read_request
     hook in http_core.c.  This removes the need to add the filter in
     multiple places and allows for an SSL module to be added much
     simpler. [Ryan Bloom]

  *) Fix a security problem that affects certain configurations of
     mod_rewrite. If the result of a RewriteRule is a filename that
     contains expansion specifiers, especially regexp backreferences
     $0..$9 and %0..%9, then it may be possible for an attacker to
     access any file on the web server. [Tony Finch]

  *) Fix a bug where errors that are detected during early request parsing
     don't produce visible HTTP error messages at the browser, because
     the core_filter wasn't present.  [Greg Ames]

  *) Provide apr_socklen_t as a portability aid. 
     [Victor  J. Orlikowski]

  *) Overhaul of dbmmanage to allow a groups arg (as in Apache 1.2)
     as well as a comment arg to the add, adduser and update cmds.
     update allows the user to clear or preserve pw/groups/comment.
     Fixed a bug in dbmmanage that prevented the check option from 
     parsing a password followed by :group... text.  Corrected the
     seed calcualation for Win32 systems, and added -lsdbm support.
     [William Rowe]

  *) Configured mod_auth_dbm to compile with sdbmlib under Win32.
     [William Rowe]

  *) Avoid a segfault when parsing .htaccess files.  An 
     uninitialized tree pointer was passed to ap_build_config().
     [Jeff Trawick]

  *) Change the way that inet_addr & inet_network are checked for
     in APR's configure process to allow BeOS BONE to correctly
     find them. With this change BeOS BONE now builds from source
     with no problems.  [David Reid]

  *) Fix a bug in apr_create_process() for Unix.  The NULL signifying
     the end of the parameters to execve() was stored in the wrong
     location, overlaying the storage beyond the newargs[] array and 
     also passing uninitialized storage to execve(), which would 
     sometimes fail with EFAULT.  [Jeff Trawick]

  *) Fix a bug parsing configuration file containers.  With a sequence
     like this in the config file

       <IfModule mod_kilroy.c>
       any stuff
       </IfModule>
       <IfModule mod_lovejoy.c>
       (blank line)
       any stuff
       </IfModule>

     the second container would be terminated at the blank line due to
     sediment in the buffer from reading the prior </IfModule> and an 
     error message would be generated for the real </IfModule> for the
     second container.  Also due to this problem, any two characters 
     could be used for "</" in the close of a container.  
     [Jeff Trawick]

  *) ap_add_filter prototype changed to remove the ctx pointer.  The
     pointer still remains in the filter structure, but it can not be
     a part of the ap_add_filter prototype.  The reason is that when
     the core uses AddFilter to add a filter to the stack it doesn't
     know how to allocate the ctx pointer, or even how much memory should
     be allocated.  The filters will have to be responsible for allocating
     the ctx memory when they need it.
     [Ryan Bloom]

  *) Add an AddFilter directive.  This directive takes a list of filters
     that should be activated for the requested resource.
     [Ryan Bloom]

  *) apr_snprintf(): Get quad format strings working on OS/390 (and perhaps
     some other platforms).  [Jeff Trawick]

  *) Modify mod_include to be a filter.  Currently, it has only been tested
     on actual files, but it should work for CGI scripts too.
     [Ryan Bloom]

  *) apr_putc(), apr_puts() for Unix: handle buffered files and interrupted
     writes.  apr_flush() for Unix: handle interrupted writes.
     [Jeff Trawick]

  *) NameVirtualHost can now take "*" as an argument instead of
     an IP address. This allows you to create a purely name-based
     virtual hosting server that does not have any IP addresses in
     the configuration file and which ignores the local address
     of any connections. PR #5595, PR #4455 [Tony Finch]

  *) Fix some compile warnings in mod_mmap_static.c
     [Mike Abbott <mja@sgi.com>]

  *) Fix chunking problem with CGI scripts.  The general problem was that
     the CGI modules were adding an EOS bucket and then the core added an
     EOS bucket.  The chunking filter finalizes the chunked response when it
     encounters an EOS bucket.  Because two EOS buckets were sent, we
     finalized the response twice.  The fix is to make sure we only send one
     EOS, by utilizing a flag in the request_rec.
     [Ryan Bloom]

  *) apr_put_os_file() now sets up the unget byte appropriately on Unix
     and Win32.  Previously, the first read from an apr_file_t set up via
     apr_put_os_file() would return a '\0'.  [Jeff Trawick]

  *) Mod_cgid now creates a single element bucket brigade, with a pipe
     bucket, instead of using BUFF's and ap_r*.
     [Ryan Bloom]

  *) APRVARS.in no longer overwrites the EXTRA_LIBS variable.
     [Mike Abbott <mja@sgi.com>]

  *) Remove ap_bopenf from buff code.  This required modifying the file_cache
     code to use APR file's directly instead of going through BUFFs.
     [Ryan Bloom]

  *) Fix compile break on some platforms for mod_mime_magic.c
     [John K. Sterling <sterling@covalent.net>]

  *) Fix merging of AddDefaultCharset directive.
     PR #5872 (1.3) [Jun Kuriyama <kuriyama@imgsrc.co.jp>]

  *) Minor revamp of the rlimit sections of code. We now test
     explicitly for setrlimit and getrlimit. Also, unixd_set_rlimit()
     is now "available" even if the platform doesn't support
     the rlimit family (it's just a noop though). [Jim Jagielski]

  *) Migrate the pre-selection of which MPM to use for specific
     platforms to hints.m4, which contains (or should contain)
     all platform specific "hints". [Jim Jagielski]

  *) Remove IOLs from Apache.  With filtering, IOLs are no longer necessary
     [Ryan Bloom]

  *) Add tables with non-string/binary values to APR.
     [Ken Coar]

  *) Fix some bad calls to ap_log_rerror() in mod_rewrite. 
     [Jeff Trawick]

  *) Update PCRE to version 3.2.  [Ryan Bloom]

  *) Change the way buckets' destroy functions are called so that
     they can be more directly used when changing the type of a
     bucket in place. [Tony Finch]

  *) Add generic support for reference-counting the resources used by
     buckets, and alter the HEAP and MMAP buckets to use it. Change
     the way buckets are initialised to support changing the type of
     buckets in place, and use it when setting aside TRANSIENT buckets.
     Change the implementation of TRANSIENT buckets so that it can be
     mostly shared with IMMORTAL buckets, which are now implemented.
     [Tony Finch]

Changes with Apache 2.0a6

  *) Add support to Apache and APR for dsos on OS/390.  [Greg Ames]

  *) Add a chunking filter to Apache.  This brings us one step closer
     to removing BUFF. [Ryan Bloom]

  *) ap_add_filter now adds filters in a LIFO fashion.  The first filter
     added to the stack is the last filter to be called.  [Ryan Bloom]

  *) Apache 2.0 has been completely documented using Scandoc.  The
     docs can be generated by running 'make docs'.  [Ryan Bloom]

  *) Add filtered I/O to Apache.  This is based on bucket brigades,
     Currently the buckets still use BUFF under the covers, but that
     should change quickly.  The only currently written filter is the
     core filter which just calls ap_bwrite.  [The Apache Group]

  *) APR locks on Unix: Let APR_LOCKALL locks work when APR isn't
     built with thread support.  [Jeff Trawick]

  *) Abort configuration if --with-layout was specified and there's
     no layout definition file.  [Ken Coar]

  *) Add support for '--with-port=n' option to configure.  [Ken Coar]

  *) Add support for extension methods for the Allow response header
     field, and an API routine for accessing r->allowed and the
     list of extension methods in a unified manner.  [Ken Coar]

  *) mod_cern_meta: fix broken file reading loop in scan_meta_file().
     [Rob Simonson <simo@us.ibm.com>]

  *) Get xlate builds working again.  The apr renaming in 2.0a5 broke
     APACHE_XLATE builds.  [Jeff Trawick]

  *) A configuration file parsing problem was fixed.  When the 
     configuration file started with an IfModule/IfDefine container, 
     only the last statement in the container would be retained.  
     [Jeff Trawick]

Changes with Apache 2.0a5

  *) Perchild is serving pages after passing them to different child
     processes.  There are still a lot of bugs, but this does work.  I
     have made requests against the same installation of Apache, and had
     different servers use different user IDs to serve the responses.
     This change moves to using socketpair instead of an AF_UNIX socket.
     [Ryan Bloom]

  *) Perchild MPM still doesn't work perfectly, but it is serving pages.
     It can't seem to pass between child processes yet, but I think we
     are closer now than before.  This moves us back to using Unix
     Domain Sockets.  [Ryan Bloom]

  *) libapr functions and types renamed with apr_ prefix.
     #include "apr_compat.h" for 1.3.x backwards compat
     [Perl]

  *) Fix problems with APR sockaddr handling on Win32.  It didn't always
     return the right information on the local socket address.
     [Gregory Nicholls <gnicholls@level8.com>]

  *) ap_recv() on Win32: Set bytes-read to 0 on error.  
     [Gregory Nicholls <gnicholls@level8.com>]

  *) Add an option to not detach from the controlling terminal without
     going into single process mode.  This allows for much easier
     debugging of the process startup code. [Ryan Bloom]

  *) ab: don't use perror() to report the failure of an APR function.
     [Jeff Trawick]

  *) Make dexter, mpmt_pthread, and perchild MPMs not destroy the
     scoreboard on graceful restarts.
     [Ryan Bloom]

  *) Fix segfault/SIGSEGV when running gzip from mod_mime_magic.c.
     An invalid ap_proc_t was passed to ap_create_process().
     [Jeff Trawick]

  *) Allow modules to register filters.  Those filters are still
     never called, but this is a step in the right direction.
     [Ryan Bloom and Greg Stein]

  *) Register the mod_cgid daemon process for cleanup so that it is
     killed at termination if it does not die when the parent gets
     SIGTERM.  This change is to fix occasional problems where the
     process stays around.  Bugs in similar logic in mod_rewrite and
     mod_include were also fixed.  [Jeff Trawick]

  *) Fix a bug in the time handling.  Basically, we were imploding a time
     in ap_parseHTTPdate, but it had bogus data in the exploded time format.
     Namely, tm_usec and tm_gmtoff were not filled out.  ap_implode_time
     uses those two fields to adjust the time value.  Because of the HTTP
     spec, both of those values can be zero'ed out safely.  This fixes
     the bug correctly.  [Ryan Bloom]

  *) Fix a couple of place in the Windows code where the wrong error
     code was being returned. [Gregory Nicholls <gnicholls@level8.com>]

  *) Fix POOL_DEBUG (at least for prefork mpm). [Dean Gaudet]

  *) Added the APR_EOL_STR macro for platform dependent differences in 
     logfiles and other raw text (such as all APR files).  Fixes logfiles
     not terminated with cr/lf sequences in Win32.  [William Rowe]

  *) Move all strings functions in APR to src/lib/apr/strings and create
     apr_strings.h for the prototypes. [Ryan Bloom]

  *) APR lock fixes: when using SysV sems, flock(), or fcntl(), be sure
     to repeat the syscall until we stop getting EINTR.  I noticed a
     related problem at termination (SIGTERM) on FreeBSD when using
     fcntl().  Apache 1.3 had these new loops too.  Also, make the flock() 
     implementation work properly with child init.  Previously, ap_lock()
     was essentially a no-op because all children were using different
     locks and thus nobody ever blocked.  [Jeff Trawick]

  *) The htdocs/ tree has been moved out of the CVS source tree into
     a separate area for easier development.  This has NO EFFECT on
     end-users or Apache installations.  [Ken Coar]

  *) Integrate the mod_dav module for WebDAV protocol handling. This
     adds the dav and dav_fs modules, the SDBM library, and additional
     XML handling utilities. [Greg Stein]

  *) Clean out obsolete names (from httpd.h) for the HTTP Status Codes
     [Greg Stein]

  *) Update the lib/expat-lite/ library (bring forward changes from
     the Apache 1.3 repository). [Greg Stein]

  *) If sizeof(long long) == sizeof(long), then prefer long in APR
     configure.in.  [Dave Hill <ddhill@zk3.dec.com>]

  *) Add ap_sendfile for Tru64 Unix.  Also, add an error message for
     machines where sendfile is detected, but nobody has written ap_sendfile.
     [Dave Hill <ddhill@zk3.dec.com>]

  *) Compile fixes in mod_mmap_static.  [Victor J. Orlikowski]

  *) ab would start up more connections than needed, then quit when the
     desired number were finished. Also fixed a logic error involving
     ab keepalives.  [Victor J. Orlikowski]

  *) WinNT: Implement non-blocking pipes with timeouts to communicate
     with CGIs. Apache 2.0a4 had non-blocking pipes but without 
     timeouts (i.e, if a timeout was specified, the pipe reverted to
     a full blocking pipe). Now the behaviour is more in line with
     Unix non-blocking pipes.
     [Bill Stoddard]

  *) WinNT: Implement accept socket reuse. Using mod_file_cache to
     cache open file handles along with accept socket reuse enables
     Apache 2.0 to serve non-keepalive requests for static files at
     3x the rate of Apache 1.3.(e.g, Apache 1.3 will serve 400 rps
     and Apache 2.0 will serve almost 1200 rps on my system).
     [Bill Stoddard]

  *) Merge mod_mmap_static function into mod_file_cache. mod_file_cache
     supports two config directives, mmapfile (same behavious as
     mod_mmap_static) and cachefile. Use the cachefile directive
     to cache open file handles. This directive only works on systems
     that have implemented the ap_sendfile API. cachefile works today
     on Windows NT, but has not been tested on any flavors of Unix.
     [Bill Stoddard]

  *) Cleanup the configuration.  With the last few changes the
     configuration process automatically:
         inherits information about how to build from APR.  Allowing
         APR to inform Apache that it should or should not use -ldl
        
         Detects which mod_cgi should be used mod_cgi or mod_cgid,
         based on the threading model

         Apache calls APR's configure process before finishing it's
         configuration processing, allowing for more information flow
         between the two.
     [Ryan Bloom]
         

  *) Change Unix and Win32 ap_setsockopt() so that APR_SO_NONBLOCK
     with non-zero argument makes the socket non-blocking.  BeOS and
     OS/2 already worked this way.  [Jeff Trawick]

  *) ap_close() now calls ap_flush() for buffered files, so write
     operations work a whole lot better on buffered files.
     [Jeff Trawick]

  *) Fix error messages issued from MPMs which explain where to change
     compiled-in limits (e.g., ThreadsPerChild, MaxClients, StartTreads).
     [Greg Ames]

  *) ap_create_pipe() now leaves pipes in blocking state.  (This helps 
     reduce the number of syscalls on Unix.)  ap_set_pipe_timeout() is
     now the way that the blocking state of a pipe is manipulated.
     ap_block_pipe() is gone.  [Jeff Trawick]

  *) Correct the problem where the only local host name that the IP stack
     can discover are 'undotted' private names.  If no fully qualified
     domain name can be identified, the default ServerName will be set to
     the machine's IP address string. A warning is always provided if the
     ServerName not specified, but assumed.  Solves PR6215 [William Rowe]

  *) Repair problems with config file processing which caused segfault
     at init when virtual hosts were defined and which caused ServerName to
     be ignored when there was no valid DNS setup.  [Jeff Trawick]

  *) Removed pointless ap_is_aborted macro function. [Roy Fielding]

  *) Add ap_sendfile implementation for AIX
     [Victor J. Orlikowski]

  *) Repair C++ compatibility in ap_config.h, apr_file_io.h, 
     apr_network_io.h, and apr_thread_proc.h.  
     [Tyler J. Brooks <tylerjbrooks@home.com>, Jeff Trawick]

  *) Bring the allocation and pool debugging code back into a working
     state.  This will need to be tested as so far it's only been used on
     BeOS. [David Reid]

  *) Change configuration command setup to be properly typesafe when in
     maintainer mode. Note that this requires a compiler that can initialise
     unions. [Ben Laurie]

  *) Turn on buffering for config file reads.  Part of this was to
     repair buffered I/O support in Unix and implement buffered
     ap_fgets() for all platforms.  [Brian Havard, Jeff Trawick]

  *) Win32: Fix problem where UTC offset was not being set correctly