Re: SeaMonkey 2.35b7
"Doug Bissett" <[email protected]>
| Newsgroups | gmane.comp.mozilla.devel.os2 |
|---|---|
| Message-ID | <[email protected]> |
On Tue, 9 Aug 2016 06:47:26 UTC, Dave Yeo <[email protected]> wrote: > Doug Bissett wrote: > > On Tue, 9 Aug 2016 03:13:36 UTC, Dave Yeo <[email protected]> > > wrote: > > > >> Doug Bissett wrote: > >>> On Mon, 8 Aug 2016 18:17:15 UTC, Steve Wendt <[email protected]> > >>> wrote: > >>> > >>>> On 8/7/2016 6:38 PM, Doug Bissett wrote: > >>>> > >>>>>> I have 256 MBs of low shared memory available. > >>>>> > >>>>> I start out with less than that, after booting (last boot it was 241 > >>>>> MB). When I load FF, I am left with about 115 MB. > >>>>> > >>>>> Well, you are a special case. > >>>> > >>>> Selective snipping to make a point - clearly you are doing something > >>>> unusual at startup. You've said it's not ClamAV, but it must be > >>>> something that you are running. > >>> > >>> Some investigation reveals that CLAMD.EXE is using about 6 MB of low > >>> shared memory. I assume that is from the "common" DLLs that it uses. > >>> Running the AOO QuickStart thing uses an additional 3 MB. XCenter > >>> seems to use about 20 MB. All of that accounts for the 29 MB that gets > >>> used between the time Above512Watch gets started, until the sytstem is > >>> ready to use. At that point it says I have about 241 MB available. If > >>> I start FF (with, or without, RUN!), it drops to about 130 MB. > >>> > >>> I will need to dig into CONFIG.SYS to see what it starts up. > >>> > >> > >> I can't remember if you have DLLBASING=OFF at the head of your > >> config.sys. > > > > Yes, I do, but I think that is part of the problem. It causes > > fragmentation of the shared memory, and that may prevent getting > > enough contiguous memory for the programs. Unfortunately, not using it > > is even worse. > > > >> I do and it gave me quite a bit more (low) shared memory. > > > > I am pretty sure that there is lots of free memory, but it isn't > > contiguous after a few starts and stops, so those huge DLLs can't > > load. > > What do you have set for VIRTUALADDRESSLIMIT? I have 3072 (needed for > linking debug xul.dll) and need to test lowering it. There is a danger > of stepping on system (hardware) memory. I am using 3072. This machine has 8 G in it. I use part of that with the QSINIT RAMDISK, but that is all above 4 G, except the driver. From past experimenting, 1024 has one set of problems, 2048 has another set of problems. 1536 has fewer problems, but I see problems from both 1024 and 2048. 3072 has problems, but not as many as 1536. > > > >> Still, there are other problems. Today I fired up SM 2.5 (Gecko 8) to > >> fix places not keeping history, this never was marked to load high and > >> Free Shared Mem showed a 100MB drop right away, things ran smooth with > >> about 150 MBs free, upon closing I saw the free memory briefly drop to > >> 100 MBs before rebounding to about 250 MBs. I also took the opportunity > >> to mark some of the requirements to load high, Fontconfig, Pango, > >> Freetype, Cairo and Pixman for now. > > > > The main reason why I haven't tried some of them, is that I am worried > > about other programs that might use them. If something tries to use > > them, and it isn't designed to use high memory, there will be trouble. > > I think as long as they're not calling the 16 API, they're fine. How would a use know? > > > If it is designed to use high memory, it will likely crash on exit. I > > have tried a few things (like the XCENTER plugins), but that doesn't > > work. > > Some of the plugins probably call the 16 API. I know the drives object does. What I see is a blank screen when the XCenter(s) start up. > > > >> Going back to SM 2.35 things ran fine, history was back to working and > >> starting SM only used a couple of MBs of low shared memory. > > > > I do notice that FF works very well, when loaded high. Even if it just > > crashed at close, it would be almost worth the effort, but when it > > takes out the whole system, about 80% of the time, it isn't worth it > > any more. I don't like to let FF run over night, or when I am not > > around. I do load AOO high, and run the Quick Start thing to keep the > > DLLs from unloading. That seems to be good, but it also uses some > > lower shared memory. > > > >> Later after > >> having shutdown TB and SM, I fired up TB first and then clicked the SM > >> icon (run!L), bang reboot, didn't even get far enough to break > >> profiles.ini. This was with about 250MBs free lower shared memory. > > > > In a case like that, you really can't tell what happened, unless you > > take a system dump, and know how to use it. Even then, I suspect that > > all it would show would be a memory overlay. > > > >> It's irritating. > >> Dave > > > > It needs to be fixed, somehow. What I find "irritating" is that I have > > watched lower shared memory step down to 2 MB, before a program crash, > > while I have almost 2000 MB of upper shared memory available. > > Unloading upper shared memory seems to be where one of the problems > > lies, so it would make sense to try to keep those DLLs, that use it, > > loaded all of the time (seems to work with AOO). > > I've found that things get funky when lower shared memory gets below 10 > MBs, though my monitoring program disagrees with others by about 10 MBs > so it is hard to say exactly. Mine gets funky when it drops below 90 meg (using Above512Watch). That doesn't show exactly what Free Shared Mem shows, and SHMEMMON is quite different. > It was interesting when I accidentally > built SM without -Zhigh-mem so memory was getting allocated low. > Couldn't even finish loading my default profile. That happens after starting and stopping FF a few times. > Understand that we allocate memory high as well as hopefully load the > DLL code and data high. Have you tested just loading the code or data > high? I have been doing both. > That's what I did (code IIRC) before getting the 106 kernel as the > 105 kernel did crash on exit here. I am using the 106 kernel. Apparently part of the problem was fixed, but there are still problems. Somebody is working on it, but I have seen no results, yet. > Hopefully the 200 kernel that Blue Lion uses will fix these issues. They are trying. Whether it will fix it, or not, remains to be seen. > > > > Other things, like AE.EXE need to be investigated, to see if they can > > be rewritten to use upper shared memory, or better yet, private > > memory. > > It probably allocates memory in the low arena. Unluckily I think it was > written in Pascal, which has no support for high memory. If written in C > or C++, it could be recompiled and linked with GCC or the latest > unreleased OW to allocate memory high. I question why it uses shared memory at all, but that may be out of the programmer's control. > > I have been looking to see if I can find a substitute program, > > but haven't found one yet. JEdit looks okay, but takes far too long to > > load. EPM doesn't seem to use much shared memory, but it will take a > > bit of getting used to. > > > > I use EPM sometimes. It also has a habit of crashing occasionally, > perhaps due to memory problems. I like it but it is a line editor so > somewhat weird. That is why I rarely use it (only to peek at locked log files, mostly). I should try TEdit. > Seems there should be lots of editors floating around as it is a usual > practice program. I haven't had time to look. > Dave -- From the eComStation of Doug Bissett dougb007 at telus dot net (Please make the obvious changes, to e-mail me)