6:19 p.m.
Whilst mucking about in the code I noticed a
possible memory leak:
1161a1162,1164
//dom:memory leak?
box_free(box);
Cheers
dom
7:06 p.m.
In message <Pine.LNX.4.44.0412221727350.18183-100000(a)tarrant.ecs.soton.ac.uk>
John-Mark Bell <jmb202(a)ecs.soton.ac.uk> wrote:
On Wed, 22 Dec 2004, Dom wrote:
>
>
> Whilst mucking about in the code I noticed a
> possible memory leak:
>
> 1161a1162,1164
> > //dom:memory leak?
> > box_free(box);
> >
Which file is this in? (I presume render/box.c? (I can't be sure as my
local copy is modified somewhat))
When generating patches/diffs, please use unified diffs, produced with
"diff -u file1.c file.c" or "cvs diff -u file.c", etc, and paste the
entire context.
The version of libxml2 on
http://netsurf.strcprstskrzkrk.co.uk/developer/ wants
to link to two symbols __nan and __huge_valf...
I've compiled a later version of libxml2 which seems
ok if that needs to be uploaded instead??
Using arbitarily compiled libraries is something we really want to get
away from on RISC OS, and why I've provided repeatable builds of
libraries with known options on my site. The only reason that NetSurf
developers have provided their own versions is that they've turned off
many options in libxml, curl, etc to make them much smaller (because
we're sadly still stick with static linking).
...which aren't present in gcc.
GCC doesn't provide very many symbols itself (mostly integer/float
conversion and division and such), so this isn't a very useful comment.
It may well have been present in Unixlib in the past before maths
functionality changed.
--
Peter Naulls - peter(a)chocky.org | http://www.chocky.org/
----------------------------------------------------------------------------
Please Reply Properly - http://www.i-hate-computers.demon.co.uk/quote.html
2:38 p.m.
On Wed, 22 Dec 2004 18:06:49 GMT, Peter Naulls <peter(a)chocky.org> wrote:
In message
<Pine.LNX.4.44.0412221727350.18183-100000(a)tarrant.ecs.soton.ac.uk>
John-Mark Bell <jmb202(a)ecs.soton.ac.uk> wrote:
>
> On Wed, 22 Dec 2004, Dom wrote:
>
> >
> >
> > Whilst mucking about in the code I noticed a
> > possible memory leak:
> >
> > 1161a1162,1164
> > > //dom:memory leak?
> > > box_free(box);
> > >
>
> Which file is this in? (I presume render/box.c? (I can't be sure as my
> local copy is modified somewhat))
When generating patches/diffs, please use unified diffs, produced with
"diff -u file1.c file.c" or "cvs diff -u file.c", etc, and paste
the
entire context.
Fair enough, tis my first time using cvs etc in anger today.
> The version of libxml2 on
> http://netsurf.strcprstskrzkrk.co.uk/developer/ wants
> to link to two symbols __nan and __huge_valf...
> I've compiled a later version of libxml2 which seems
> ok if that needs to be uploaded instead??
Using arbitarily compiled libraries is something we really want to get
away from on RISC OS, and why I've provided repeatable builds of
libraries with known options on my site. The only reason that NetSurf
developers have provided their own versions is that they've turned off
many options in libxml, curl, etc to make them much smaller (because
we're sadly still stick with static linking).
Fair enough, that is why I wanted to sort this out, which version
_should_ people be using though? I'll use the version from riscos.info
from now on which seems to be smaller to the one on netsurf..../developer?
Can we use this one as standard and copy to the developer page? It is
good to have a set of "standard" libraries against which to code so
we're all on the same page
> ...which aren't present in gcc.
GCC doesn't provide very many symbols itself (mostly integer/float
conversion and division and such), so this isn't a very useful comment.
It may well have been present in Unixlib in the past before maths
functionality changed.
By GCC I meant GCC/Unixlib, I'm sorry if this lack of clarity has so
offended you. It does look like it came from Unixlib at some point
it's now declared as static though.
Dom
3:59 p.m.
On Thu, 23 Dec 2004, Dom wrote:
[libxml2]
Fair enough, that is why I wanted to sort this out, which version
_should_ people be using though? I'll use the version from riscos.info
from now on which seems to be smaller to the one on netsurf..../developer?
Not by my reckoning:
netsurf.strcprstskrzkrk.co.uk/developer/libxml2.zip 318k
riscos.info/downloads/unix-libraries/libxml2-2.6.11.zip 599k
Note also that the NetSurf one is version 2.6.16
Can we use this one as standard and copy to the developer page? It
is
good to have a set of "standard" libraries against which to code so
we're all on the same page
Doing that would add a fairly large amount to the size of the NetSurf
binary.
John.
4:18 p.m.
On 23 Dec 2004 John-Mark Bell <jmb202(a)ecs.soton.ac.uk> wrote:
On Thu, 23 Dec 2004, Dom wrote:
> Can we use this one as standard and copy to the developer page? It is
> good to have a set of "standard" libraries against which to code so
> we're all on the same page
Doing that would add a fairly large amount to the size of the NetSurf
binary.
I take it shaving a few K off the binary is driven by the NC builds? And also
you wouldn't want to build the desktop and NCs against different libraries?
Cheers
---Dave
--
____________________________________________________________________________
David J. Ruck Phone: +44- (0)7974 108301 Email: druck(a)druck.org.uk
____________________________________________________________________________
4:48 p.m.
On Thu, 23 Dec 2004, David J. Ruck wrote:
On 23 Dec 2004 John-Mark Bell <jmb202(a)ecs.soton.ac.uk> wrote:
> On Thu, 23 Dec 2004, Dom wrote:
> > Can we use this one as standard and copy to the developer page? It is
> > good to have a set of "standard" libraries against which to code so
> > we're all on the same page
>
> Doing that would add a fairly large amount to the size of the NetSurf
> binary.
I take it shaving a few K off the binary is driven by the NC builds?
No, and it's more likely to be a few hundred k. In fact, the libraries I
link against render a binary some 6-700 k larger than the official
download (and these have some functionality turned off too, just not as
much as those on the autobuilder). Thus, linking against vanilla libraries
would probably add at least a meg to the size of the !RunImage, which
really is excessive, given that it's functionality we don't use (and most
likely won't ever use). Of course, we could always turn NetSurf into some
20Meg beast, if you'd prefer ;)
And also you wouldn't want to build the desktop and NCs against
different libraries?
It reduces hassle, yes (although I doubt the NC builds work at all, any
more).
John.
5:21 p.m.
On 23 Dec 2004 John-Mark Bell <jmb202(a)ecs.soton.ac.uk> wrote:
On Thu, 23 Dec 2004, David J. Ruck wrote:
> I take it shaving a few K off the binary is driven by the NC builds?
No, and it's more likely to be a few hundred k. In fact, the libraries I
link against render a binary some 6-700 k larger than the official
download (and these have some functionality turned off too, just not as
much as those on the autobuilder). Thus, linking against vanilla libraries
would probably add at least a meg to the size of the !RunImage, which
really is excessive, given that it's functionality we don't use (and most
likely won't ever use). Of course, we could always turn NetSurf into some
20Meg beast, if you'd prefer ;)
Well with a 512MB Iyonix, and a 194MB Risc PC, I'm not bothered about another
MB or two on the binary, especially as I'll be using at least 32MB of cache
when running.
It may still be of concern to some tight git with a tiny amount of RAM,
so it may be worth paring down the libraries for a release build. But I'm
thinking it may save you some time in development if you use the full
libraries for the test builds.
Cheers
---Dave
--
____________________________________________________________________________
David J. Ruck Phone: +44- (0)7974 108301 Email: druck(a)druck.org.uk
____________________________________________________________________________
5:27 p.m.
In message <675adf214d.druck(a)druck.freeuk.net>
"David J. Ruck" <druck(a)druck.org.uk> wrote:
It may still be of concern to some tight git with a tiny amount of
RAM,
so it may be worth paring down the libraries for a release build. But I'm
thinking it may save you some time in development if you use the full
libraries for the test builds.
Cool! so you're gonna buy memory upgrades for all those who can't afford it
themselves?
Tell me when you have got me one and I'll email you my address ;)
--
5:50 p.m.
On 23 Dec 2004 Andrew Duffell <andrewduffell(a)yahoo.co.uk> wrote:
In message <675adf214d.druck(a)druck.freeuk.net>
"David J. Ruck" <druck(a)druck.org.uk> wrote:
> It may still be of concern to some tight git with a tiny amount of RAM,
> so it may be worth paring down the libraries for a release build. But I'm
> thinking it may save you some time in development if you use the full
> libraries for the test builds.
Cool! so you're gonna buy memory upgrades for all those who can't afford it
themselves? Tell me when you have got me one and I'll email you my address
;)
There is probably a couple of 16MB or 32MB sticks gathering dust here
somewhere. But I think you'll find the active RISC OS users bothering to
download test builds of software, have more than RAM.
Cheers
---Dave
--
____________________________________________________________________________
David J. Ruck Phone: +44- (0)7974 108301 Email: druck(a)druck.org.uk
____________________________________________________________________________
7:08 p.m.
In message <3ce5df214d.yahoo(a)hotpop.com>
Andrew Duffell <andrewduffell(a)yahoo.co.uk> wrote:
In message <675adf214d.druck(a)druck.freeuk.net>
"David J. Ruck" <druck(a)druck.org.uk> wrote:
> It may still be of concern to some tight git with a tiny amount of RAM,
> so it may be worth paring down the libraries for a release build. But I'm
> thinking it may save you some time in development if you use the full
> libraries for the test builds.
Cool! so you're gonna buy memory upgrades for all those who can't afford it
themselves?
Tell me when you have got me one and I'll email you my address ;)
About 12-15GBP is what they seem to go for on ebay for a single 64mb
simm.
Might even be less now with Christmas so close (prices also seem to
plummet in the aftermath of Christmas too).
Cheers!
--
Graham
The Main Control Room - www.thedeathzone.free-online.co.uk
Deathzone Emulation - www.thedeathzone.free-online.co.uk/emulation
12:28 a.m.
In message <f128e9214d.TheDoctor(a)thedeathzone.free-online.co.uk>
Graham <thedoctor(a)thedeathzone.free-online.co.uk> wrote:
In message <3ce5df214d.yahoo(a)hotpop.com>
Andrew Duffell <andrewduffell(a)yahoo.co.uk> wrote:
>
> In message <675adf214d.druck(a)druck.freeuk.net>
> "David J. Ruck" <druck(a)druck.org.uk> wrote:
> > It may still be of concern to some tight git with a tiny amount of RAM,
> > so it may be worth paring down the libraries for a release build. But I'm
> > thinking it may save you some time in development if you use the full
> > libraries for the test builds.
>
> Cool! so you're gonna buy memory upgrades for all those who can't afford it
> themselves?
> Tell me when you have got me one and I'll email you my address ;)
>
About 12-15GBP is what they seem to go for on ebay for a single 64mb
simm.
Might even be less now with Christmas so close (prices also seem to
plummet in the aftermath of Christmas too).
That wasn't the point though. I actually have plenty or RAM in *my* computer.
--
12:21 p.m.
In message <947f06224d.yahoo(a)hotpop.com>
Andrew Duffell <andrewduffell(a)yahoo.co.uk> wrote:
In message <f128e9214d.TheDoctor(a)thedeathzone.free-online.co.uk>
Graham <thedoctor(a)thedeathzone.free-online.co.uk> wrote:
>
> In message <3ce5df214d.yahoo(a)hotpop.com>
> Andrew Duffell <andrewduffell(a)yahoo.co.uk> wrote:
>
> >
> > In message <675adf214d.druck(a)druck.freeuk.net>
> > "David J. Ruck" <druck(a)druck.org.uk> wrote:
> > > It may still be of concern to some tight git with a tiny amount of RAM,
> > > so it may be worth paring down the libraries for a release build. But
I'm
> > > thinking it may save you some time in development if you use the full
> > > libraries for the test builds.
> >
> > Cool! so you're gonna buy memory upgrades for all those who can't
afford it
> > themselves?
> > Tell me when you have got me one and I'll email you my address ;)
> >
> About 12-15GBP is what they seem to go for on ebay for a single 64mb
> simm.
> Might even be less now with Christmas so close (prices also seem to
> plummet in the aftermath of Christmas too).
>
That wasn't the point though. I actually have plenty or RAM in *my* computer.
Perhaps then I should ask:
Is there anyone who actually uses their Acorn for web browsing that
cannot afford £15 for a memory upgrade?
I can't imagine there are many.
Cheers!
--
Graham
The Main Control Room - www.thedeathzone.free-online.co.uk
Deathzone Emulation - www.thedeathzone.free-online.co.uk/emulation
5:39 p.m.
On Thu, 23 Dec 2004, David J. Ruck wrote:
On 23 Dec 2004 John-Mark Bell <jmb202(a)ecs.soton.ac.uk> wrote:
> On Thu, 23 Dec 2004, David J. Ruck wrote:
> > I take it shaving a few K off the binary is driven by the NC builds?
>
> No, and it's more likely to be a few hundred k. In fact, the libraries I
> link against render a binary some 6-700 k larger than the official
> download (and these have some functionality turned off too, just not as
> much as those on the autobuilder). Thus, linking against vanilla libraries
> would probably add at least a meg to the size of the !RunImage, which
> really is excessive, given that it's functionality we don't use (and most
> likely won't ever use). Of course, we could always turn NetSurf into some
> 20Meg beast, if you'd prefer ;)
Well with a 512MB Iyonix, and a 194MB Risc PC, I'm not bothered about another
MB or two on the binary, especially as I'll be using at least 32MB of cache
when running.
;)
It may still be of concern to some tight git with a tiny amount of
RAM,
so it may be worth paring down the libraries for a release build. But I'm
thinking it may save you some time in development if you use the full
libraries for the test builds.
Building new versions of the libraries takes very little time indeed, as
it's all cross-compiled so it's just a case of configuring the package
then making it. That can then be reduced to invoking (a single) shell
script(s), which contain(s) the relevant commands.
For example, this is the config line I use for curl:
/home/riscos/env/ro-config --enable-http --enable-ftp --enable-file
--disable-ldap --disable-dict --enable-gopher --disable-telnet
--disable-manual --disable-ipv6 --enable-nonblocking --enable-thread
--enable-ares=/home/riscos/env/lib --disable-debug
--with-random=/dev/urandom --with-ssl=/home/riscos/env/lib
--with-zlib=/home/riscos/env/lib --without-libidn
(which I've just grabbed out of the relevant script). As it happens with
the latest release of the porting tools (made yesterday), this can be
simplified further as the ro-config script handles the random and ssl
definitions itself. I dare say it could be modified to specify the zlib
path, too (and this may well be sensible as zlib is commonly used).
John.
12:16 a.m.
On Wednesday 22 December 2004 17:19, Dom wrote:
Whilst mucking about in the code I noticed a
possible memory leak:
1161a1162,1164
> //dom:memory leak?
> box_free(box);
I'm currently making various improvements to box.c, including fixing leaks, so
this will probably be fixed soon.
James
6667
days inactive
6669
days old
netsurf-users@netsurf-browser.org
14 comments
7 participants
participants (7)
-
Andrew Duffell -
David J. Ruck -
Dom -
Graham -
James Bursa -
John-Mark Bell -
Peter Naulls