Re: lynx-dev dev.18 patch: USE_PRETTYSRC, lynx.cfg

[Leonid Pauzner]

15-Jan-2000 12:21 Doug Kaufman wrote:
> On Sat, 15 Jan 2000, Leonid Pauzner wrote:

>> Well, now I think -O1 would be a compromize (it does have a difference
>> over the case without -O at all). Tom, please made respective changes,
>> I also tweak INSTALLATION file.

> As far as I can tell from the gcc info files, there is no "O1" flag.
No, try "info gcc": -O1 is the synonym of -O
at least for gcc 2.952 which I have handy.

> There is "-O" and "-O2". "-O3" generally is of no benefit under DJGPP
> and sometimes gives inferior performance (according to messages
> from the usual experts in comp.os.msdos.djgpp). I would be in favor
> of leaving the makefile optimized at "-O2". Anyone having trouble
> compiling can always decrease optimization. That seems more reasonable
> to me than leaving the standard makefile without full optimization.
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^
Why without, I propose -O1 which seems enough.

>> -    If using DJGPP 2.02 and GCC 2.8.1, you may not need to "stubedit"
>> -    cc1.exe if you use -O2 optimization, since cc1.exe comes with
>> -    a 1536K stack. To compile with -O3 optimization, the stub
>> -    needs to be edited to give a larger stack. To do this go into
>> -    djgpp\lib\gcc-lib\djgpp\2.81 and either type the command:
>> +    the C compiler and its libraries (see readme.1st from DJGPP 
>> distribution).

> I know that this is in the current file, but is it really true that
> you can compile with -O2 without enlarging the stack? I would have
That was you, who add this passage about 2.8.1 to INSTALLATION file.

> to go back to the default stack on my system to check this. My
> recollection is that the stack is larger, but so is the requirement. I
> think I had to enlarge the stack when I switched to 2.8.1, but I can't
> remember if that was with -O2 or -O3


> The current versions are DJGPP 2.03 and GCC 2.95.2. I haven't updated
> yet, but if we update the INSTALLATION file, we should probably make
> it current. Has anyone tested 2.03 with 2.95.2 to see what changes in
> stack and buffer are really necessary to compile lynx for DOS with
> various optimization levels?

I have updated to the recent version (no problem; G++ port much improved
for STL and StandardC++ conformance, that was my reason to update the
compiler). Just now I rebuilt lynx with -O3 on your request: no problem
(DJGPP 2.03 and GCC 2.95.2 just out of the box).

On my P100 32Mb RAM it stays significant time at SGML.c, GridText.c and
HTML.c and overall compilation time enlarged in factor ~4 compared with
-O1. Probably OK when you are preparing a public distribution but not
the best idea when you are doing development or playing with your own
copy (there is no multitasking on DOS anyway, so no benefits from a tiny
optimization nor possiblity to do something else when compile the
program). Well, I decrease optimization level because doing my
development on 386DX40 8Mb RAM machine till recently (the slowdown
factor was much larger because of *a lot* paging and disk activities,
and I have not drink coffee so long).



>                        Doug

> __
> Doug Kaufman
> Internet: address@hidden