To: sandmann AT clio DOT rice DOT edu, djgpp-workers AT delorie DOT com
Cc: eliz AT is DOT elta DOT co DOT il
References: <10210150631 DOT AA20605 AT clio DOT rice DOT edu>
Message-Id: <2.7.9.20QBT.H40UO8@pauzner.dnttm.ru>
From: "Leonid Pauzner" <uue AT pauzner DOT dnttm DOT ru>
Date: Tue, 15 Oct 2002 16:00:08 +0400 (MSD)
X-Mailer: dMail [Demos Mail for DOS v2.7.9]
Subject: Re: libc' getenv optimization (patch)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Reply-To: djgpp-workers AT delorie DOT com

15-Oct-2002 01:28 you wrote:
>> perhaps a realistic estimates is around 3-10 times for others.
>> Preparing procedure set_hash_env() will effectively slowdown putenv() a bit,
>> but we putenv less frequently than getenv, right?

> If we cache getenv() properly in the tzset(), we'll only call it once.
Agree. That was my original point, but IZ resist.

> So there's the chance that you call putenv() more frequently (in
> initialization) than getenv() at runtime.

Sorry, my wording was incorrect:
the optimization procedure is called inside getenv(),
but only if there was an environ change since previous getenv() call.

(according to profile, load_environment_file mixes putenv and getenv calls.)
-----------------------------------------------
                                                 <spontaneous>
[187]    0.0    0.00    0.00                 __crt0_load_environment_file [187]
                0.00    0.00       8/10          __putenv [188]
                0.00    0.00      13/71          getenv [194]
                0.00    0.00       1/171757      malloc [33]
                0.00    0.00       1/137564      free [41]
-----------------------------------------------

> With the very sparse distribution of letters, I think it would be much
> more space efficient to mask the lower 4 or 5 bits for the buckets
> (16 or 32 should be fine, 256 is overkill).  This also would speed up
> the initial setup.

Indeed...

> Here's my proposed patch for ctime.c:

yes, I saw this also.

> *** ctime.bak   Sun Jun  9 13:20:22 2002
> --- ctime.c     Tue Oct 15 01:24:06 2002
> *************** tzset(void)
> *** 920,926 ****

>     /* If environ didn't changed since last time, don't waste time
>        looking at $TZ.  */
> !   if (lcl_is_set > 0 && __environ_changed == last_env_changed)
>       return;

>     /* If environ did change, but $TZ wasn't changed since last time we
> --- 920,926 ----

>     /* If environ didn't changed since last time, don't waste time
>        looking at $TZ.  */
> !   if (lcl_is_set != 0 && __environ_changed == last_env_changed)
>       return;

>     /* If environ did change, but $TZ wasn't changed since last time we