Hi Christopher,
I am not sure if this is necessarily the right way to go for a
library, esp if it can impact backwards compatibility for
bufferevents. As for reducing the size of the library, do you really
think that 30K make a difference these days?
Niels.
On 9/26/07, Christopher Layne <[EMAIL PROTECTED]> wrote:
> $ ./configure --help
> [...]
> Optional Features:
> [...]
> --enable-dns build with support for dns layer [default=yes]
> --enable-http build with support for http layer [default=yes]
> --enable-bevents build with support for buffer events [default=yes]
>
> Changes:
>
> 1. This required me to move buffer_* and evbuffer_* function declarations
> that were in event.h to a new file, evbuffer.h. Also, it looks like
> buffer.c contains ev_buffer_* functions and evbuffer.c contains
> buffer_* functions. This was like this before so I just tried to keep
> things consistent overall when creating a new header file and went with
> evbuffer.h. The flip-flop can be changed later. Also added evbuffer.h include
> to various http and bufferevents specific modules/headers that required them.
>
> 2. This required creating test/regress_buffer.c and moving bevents
> regress specific code into that module. Calls to higher level regress
> suite collection functions are conditionally ifdef'd via HAVE_EVENT_DNS,
> HAVE_EVENT_HTTP, etc. (which configure takes care of).
>
> 3. In both cases of above, the actual conditional building of modules
> is handled via ifdef's within Makefile.am that configure takes care of.
>
> 4. Modified configure.in to not bother doing Fortran, C++, or ObjC specific
> tests. Libtool likes to try it's own checks as well, but I just redefined
> those with simple null macros.
>
> 5. Throughout this all: substantially increased my derision for GNU auto*
> tools.
>
> Pros:
>
> 1. Significantly smaller libevent library footprint but the ability to
> include everything normally.
> 2. Default, nothing changes, all API layers are enabled per normal.
> 3. Couple of useful macros added to configure.in which should make adding
> any other enable/disable and build checks quicker.
>
> Cons:
>
> 1. http code is highly dependent on bufferevents code. As such, one cannot
> use "--enable-bufferevents=no" w/ "--enable-http=no". If bufferevent code
> is disabled, http code must be disabled.
> 2. rpc code (which AFAICT is regress testing specific) is tied with http
> code and shares the same characteristics of above.
>
> Builds+size+make verify checks:
>
> $ set=( "--disable-none" \
> "--disable-dns" \
> "--disable-http" \
> "--disable-dns --disable-http" \
> "--disable-bevents --disable-http" \
> "--disable-bevents --disable-http --disable-dns" ); \
> a=${#set[*]}; i=0; while [[ $i -lt $a ]]; do f="${set[i++]}"; echo $f; \
> (./configure $f && make clean && make) 1>/tmp/error.out 2>&1 \
> || (cat /tmp/error.out && break); \
> size .libs/libevent.so; \
> make verify 2>&1 | egrep FAIL && break; \
> printf "\n"; done
>
> --disable-none
> text data bss dec hex filename
> 67115 1120 6904 75139 12583 .libs/libevent.so
>
> --disable-dns
> text data bss dec hex filename
> 48804 1040 6328 56172 db6c .libs/libevent.so
>
> --disable-http
> text data bss dec hex filename
> 45589 840 708 47137 b821 .libs/libevent.so
>
> --disable-dns --disable-http
> text data bss dec hex filename
> 26603 728 132 27463 6b47 .libs/libevent.so
>
> --disable-bevents --disable-http
> text data bss dec hex filename
> 37882 744 704 39330 99a2 .libs/libevent.so
>
> --disable-bevents --disable-http --disable-dns
> text data bss dec hex filename
> 18969 628 128 19725 4d0d .libs/libevent.so
>
>
> This patch is 50k, but is mostly just -+ of moving buffer-specific code from
> one file to another and the such. Attached as bzip2 here, and also raw at
> this url: http://www.anodized.com/~clayne/libevent.2007092600.diff
>
> -cl
>
> _______________________________________________
> Libevent-users mailing list
> Libevent-users@monkey.org
> http://monkey.org/mailman/listinfo/libevent-users
>
>
>
_______________________________________________
Libevent-users mailing list
Libevent-users@monkey.org
http://monkey.org/mailman/listinfo/libevent-users
