DOS ain't dead

FPC 2.6.2 has been released!

FPC 2.6.2 a pointer release from the 2.6.x series.

Downloads:
http://freepascal.org/down/i386/go32v2-netherlands.var

P.s. the CHMs for the textmode IDE haven't updated yet. Something for the weekend....

Hallo Beste vriend (as recently seen in my Spam folder, heheh)

> FPC 2.6.2 has been released!
>
> FPC 2.6.2 a pointer release from the 2.6.x series.

inc(x,2); (* *)

> Downloads:
> http://freepascal.org/down/i386/go32v2-netherlands.var
>
> P.s. the CHMs for the textmode IDE haven't updated yet. Something for the
> weekend....

I've been preoccupied with the flu this past week, ugh. So I may be slow in replying. (Maybe that's a good thing??) Anything in particular you want done or in what order? Announce for FreeDOS? Mirror to iBiblio? Make (newer) FDNPKG package? Or just general tests from the real experts (Laaca, Robert, et al.)?

> > Downloads:
> > http://freepascal.org/down/i386/go32v2-netherlands.var
> >
> > P.s. the CHMs for the textmode IDE haven't updated yet. Something for
> the
> > weekend....
>
> I've been preoccupied with the flu this past week, ugh. So I may be slow in
> replying. (Maybe that's a good thing??) Anything in particular you want
> done or in what order? Announce for FreeDOS? Mirror to iBiblio?

I don't think mirrors are necessary, FPC is already mirrored on sourceforge.

I usually mail J. Hall myself, but haven't yet, if you can forward him the release message, that would be nice. (and I won't have to dig up the email address :)

That reminds me that I also still have to mail the FreeBSD porter.

> Make
> (newer) FDNPKG package?

I don't even know what that is

> Or just general tests from the real experts (Laaca,
> Robert, et al.)?

Users in general, and contributors specially indeed.

> > I've been preoccupied with the flu this past week, ugh. So I may be slow
> in
> > replying. (Maybe that's a good thing??) Anything in particular you want
> > done or in what order? Announce for FreeDOS? Mirror to iBiblio?
>
> I don't think mirrors are necessary, FPC is already mirrored on
> sourceforge.

So is FreeDOS, and I agree SF.net is pretty reliably robust, but usually FD likes to keep necessary tools mirrored locally too (at least in theory). Though yes, I admit, there isn't officially a lot of Pascal (or FPC-specific) stuff in FD, but some does exist.

> I usually mail J. Hall myself, but haven't yet, if you can forward him the
> release message, that would be nice. (and I won't have to dig up the email
> address :)

Sure, but I'm not sure exactly which bit the "release message" actually is. Your post here was pretty minimal! But I'll be sure to tell him (though likely he'll only post a news update, esp. if I don't beat him to it).

> > Make (newer) FDNPKG package?
>
> I don't even know what that is

The new "Network" (packet-driver aware) Packager by Mateusz Viste (C, DJGPP + Watt-32). IIRC, I was emailing you and him both a while back (maybe indirectly, all I remember offhand was the FPC installer DiskFree bug). IIRC, I made a newer "package" (.ZIP with special layout) with 2.6.0 instead of old 2.0.4.

> > Or just general tests from the real experts (Laaca,
> > Robert, et al.)?
>
> Users in general, and contributors specially indeed.

Perhaps just a general email thread on freedos-user or freedos-devel would be best.

P.S. I should probably email CWS too, as it uses his CWSDPMI r7 (he's too busy to finish up fixes, but he did some minor hacking over the holidays). Though I have no good examples of memory testing for him. Perhaps I need to whip up something (GetMem, FreeMem ??).

Emailed freedos-devel (and assume Jim sees it there) about the release. Also emailed CWS directly.

Mirrored to iBiblio (dos262 and dos262full). Will have to make newer FDNPKG one of these days.

Updated FreeDOS' online .LSM to say 2.6.2.

Seems installer was updated (and now w/ srcs) but haven't tested whether it fixes the FAT32 bug or not. (Need to try on my main desktop's FAT32 partition.) For other attempt (install to FAT16 RAM disk), it does seem to (harmlessly) whine if it can't find "ufppkgdos.zip" (though it's actually named ufppkdos.zip). But works okay.

IDE shelling out proxy bug seems fixed, but something else weird is going on. Seems restarting the IDE fails on second try. Reinstalling [sic] fixes it for one more run, but .... Not sure what's going on here.

Not sure about that bugtracker re: COFF reloc limits. That's been fixed in DJGPP ports of BinUtils 2.23.1 (thanks to Juan G.), is someone (Sergei G.?) planning internal linker for GO32V2?

P.S. 2.7.1 acts like it was "Released 2011-08-03", but that's never been an official release, has it? All I see are snapshots, e.g. last one (GO32V2, at least) is on FPC FTP dated from January 17, 2013. There are 20x more fixes for 2.7.1 than 2.6.[12]. Yes, I know you have different branches and release schedules. Just saying, it still seems (barely) funny.

> Not sure about that bugtracker re: COFF reloc limits. That's been fixed in
> DJGPP ports of BinUtils 2.23.1 (thanks to Juan G.), is someone (Sergei G.?)
> planning internal linker for GO32V2?

Afaik he was working on an ELF linker, but I haven't seen commits in the last month on this subject.

> P.S. 2.7.1 acts like it was "Released 2011-08-03", but that's never been an
> official release, has it? All I see are snapshots, e.g. last one (GO32V2,
> at least) is on FPC FTP dated from January 17, 2013. There are 20x more
> fixes for 2.7.1 than 2.6.[12]. Yes, I know you have different branches and
> release schedules. Just saying, it still seems (barely) funny.

2.7.1 is the version in trunk. It is a moving version, probably what you got is some snapshot, nothing more.

> > Not sure about that bugtracker re: COFF reloc limits. That's been fixed
> in
> > DJGPP ports of BinUtils 2.23.1 (thanks to Juan G.), is someone (Sergei
> G.?)
> > planning internal linker for GO32V2?
>
> Afaik he was working on an ELF linker, but I haven't seen commits in the
> last month on this subject.

Here's what I meant (23752):

"(2013-01-26) The issue was caused by more than 65535 relocations per object section. Fixed for PECOFF targets (e.g. Windows). Not fixable for DJCOFF targets (e.g. go32v2) due to object file format limitation, for these targets compiler now prints a correct error message."

> > P.S. 2.7.1 acts like it was "Released 2011-08-03", but that's never been
> an
> > official release, has it? All I see are snapshots, e.g. last one
> (GO32V2,
> > at least) is on FPC FTP dated from January 17, 2013. There are 20x more
> > fixes for 2.7.1 than 2.6.[12]. Yes, I know you have different branches
> and
> > release schedules. Just saying, it still seems (barely) funny.
>
> 2.7.1 is the version in trunk. It is a moving version, probably what you
> got is some snapshot, nothing more.

I know all that, but I really was wondering why there is any bother with 2.6.2 at all when the 2.7.1 line has so many more fixes (that weren't backported). Plus the use of "released" sounded weird. But I guess it's just the way you guys do things. Presumably once it stabilizes, it will become 2.8.0 or whatever.

> ... Or just general tests from the real experts (Laaca,
> Robert, et al.)?

Don't expect anything from me.

---
Forum admin

> > > DJGPP ports of BinUtils 2.23.1 (thanks to Juan G.), is someone (Sergei
> > G.?)
> > > planning internal linker for GO32V2?
> >
> > Afaik he was working on an ELF linker, but I haven't seen commits in the
> > last month on this subject.
>
> Here's what I meant
> (23752):
>
> "(2013-01-26) The issue was caused by more than 65535 relocations per
> object section. Fixed for PECOFF targets (e.g. Windows). Not fixable for
> DJCOFF targets (e.g. go32v2) due to object file format limitation, for
> these targets compiler now prints a correct error message."

As said, I've no info on if he really working on it. Could be that he keeps an eye out for a future plan.

> I know all that, but I really was wondering why there is any bother with
> 2.6.2 at all when the 2.7.1 line has so many more fixes (that weren't
> backported).

The question is how many more problems have been introduced Moreover currently 2.7.x contains unfinished business in the unicode domain.

> Plus the use of "released" sounded weird. But I guess it's
> just the way you guys do things. Presumably once it stabilizes, it will
> become 2.8.0 or whatever.

Yup. And it is not weird. Nearly all products work that way. A branch for open development (2.7.x), and a stable branch (2.6.x in our case)

The point was more that in the trunk/unstable case, don't read too much into building dates. There are commits nearly every day.

> Will have to make newer FDNPKG one of these days.

TODO (though I wonder if waiting until 2.7.0 would be better).

> Seems installer was updated (and now w/ srcs) but haven't tested whether it
> fixes the FAT32 bug or not.

Still haven't tried installing "full" (with or without DOSLFN) yet.

> IDE shelling out proxy bug seems fixed, but something else weird is going
> on. Seems restarting the IDE fails on second try. Reinstalling [sic] fixes
> it for one more run, but .... Not sure what's going on here.

Just running and aborting INSTALL.EXE seems to "fix" FP.EXE for one instance, but then it borks again after it runs. At one run, I noticed that EAX had the ASCII value of "ENIL" ("LINE" backwards), but that's as close a guess (and useless) as I can gather right now. (I suppose running external GDB on it should work???)

Various other minor issues (mostly versions, e.g. FP.EXE uses GDB 7.3 internally but FPC still ships 6.11 externally too, which is probably a pain for mirroring huge sources, latest DJGPP has 7.5.1).

> All I see are snapshots, e.g. last one (GO32V2, at least) is
> on FPC FTP dated from January 17, 2013.

Newest GO32V2 snapshot is partially truncated / incomplete (no CDS in ZIP), so I couldn't test.

Scott Moore's website is back online (thankfully). I was worried there for a while. (I really need to finally package up P5 for DJGPP on iBiblio, presumably without the other OS cruft.)

Mirrored old LZMA 4.42 conversion to FPC (and my small patch) to iBiblio. Also mirrored some other stuff (FH's newdelay.pas, CB's txtfiles, your FPC-patched txtfiles). Always random other stuff to possibly mirror, but I don't know the licensing (doh), which is always such an annoying obstacle. :-/

Today I quickly tried current 2.6.2 version.
I can't recommend it at all for DOS programmers.

1)
There still remains the old pain of all 2.x.x FPC line for DOS - the bug (18113) in generating the alternate segment prefic code in generated assembler and machine code.
It is in all OS targets but in windows, linux etc. are the OS services invoked via DS selector, in DOS it is via FS selector. So it occurs only in DOS.
This bug is already fixed in 2.7.1 but the fix was not backtraced into 2.6.x line. I don't understand why.

2)
Again occured weird problems with environment settings when shelling into DOS.
This time it will be harder to trace. Last time it occured always but now I saw this problem only under Windows 98. In FreeDOS I haven't noticed the problem but very probably exist also there. To be short, under W98 it cuts some environment variables. Look here (done in IDE --> command shell, then typed "set")
D:\WINDOWS\COMMAND.COM
P=D:\WINDOWS\TEMP
PROMPT=$p$g
winbootdir=D:\WINDOWS
COMSPEC=D:\WINDOWS\COMMAND.COM
CTSYN=D:\WINDOWS
PATH=D:\WINDOWS;D:\WINDOWS\COMMAND;C:\BAT;C:\KOMP
windir=D:\WINDOWS
BLASTER=A220 I5 D1 H5 P330 T6


But should be (i.e. if I type "set" not in IDE session)

TMP=D:\WINDOWS\TEMP
TEMP=D:\WINDOWS\TEMP
PROMPT=$p$g
winbootdir=D:\WINDOWS
COMSPEC=D:\WINDOWS\COMMAND.COM
CTSYN=D:\WINDOWS
PATH=D:\WINDOWS;D:\WINDOWS\COMMAND;C:\BAT;C:\KOMP
windir=D:\WINDOWS
BLASTER=A220 I5 D1 H5 P330 T6



3)
And the very new problem is the fatal IDE unstability. IDE almost always starts only first time. I start the IDE, compile and run the program (f.e. Hello.pas) and close IDE. But when I want to start IDE again it crashes with this error message:

Exiting due to signal SIGSEGV
FPC General Protection Fault at eip=006EDE95
eax=90909090 ebx=0000002A ecx=00000088 edx=00000000 esi=008F11F8 edi=0083B504
ebp=008C0CA0 esp=008C0C90 program=C:\TP\FPC262\BIN\GO32V2\FP.EXE
cs: sel=00A7 base=8470A000 limit=008FFFFF
ds: sel=00AF base=8470A000 limit=008FFFFF
es: sel=00AF base=8470A000 limit=008FFFFF
fs: sel=00D7 base=00000000 limit=0010FFFF
gs: sel=00D7 base=00000000 limit=0010FFFF
ss: sel=00AF base=8470A000 limit=008FFFFF

Call frame traceback EIPs:
$006EDE95
$005BD905
$005BCD84
$005BC83D
$005BC7F8
$00308AF8
$0030B9F4
$0002DB50
$0002DC5A
$00002040


It occurs in both my systems - in FreeDOS and in Windows 98. To be able to start IDE again in FreeDOS I have to reboot the computer. Under W98 is enough to restart the DOS command prompt.

4)
I don't know if it has some meaning but why in the "Memory sizes" in IDE is "Stack size: 0" and "Heap size: 0" ?

So, unfortunately 2.6.2 is a bad version for DOS. The unability to restart the IDE is the killer bug.

I recomend to stay with version 2.4.2 or, of course, with 1.0.10

---
DOS-u-akbar!

Hi,

> Today I quickly tried current 2.6.2 version.
> I can't recommend it at all for DOS programmers.

These don't seem like dealbreakers to me. There are always bugs, and hopefully the workarounds are "good enough" if or until a proper fix is found.

> 1)
> There still remains the old pain of all 2.x.x FPC line for DOS - the bug
> (18113) in generating the alternate segment prefic code in generated
> assembler and machine code.

I know it's a pain, but try to separate out these particular bits of code and build that without the peephole optimizer. IIRC, that was the main problem. (Isn't there also a pragma to turn off optimization?)

> This bug is already fixed in 2.7.1 but the fix was not backtraced into
> 2.6.x line. I don't understand why.

I don't either. 2.6.2 is "stable" but still has bugs that should have been fixed. I guess they're overworked (volunteers, after all) or just very optimistic that 2.7.1 will come out fairly soon.

> 2)
> Again occured weird problems with environment settings when shelling into
> DOS. Under W98 it cuts some environment variables.

The only difference I see here is TMP and TEMP missing. Maybe that's intentional for some reason, dunno.

> 3)
> And the very new problem is the fatal IDE unstability. IDE almost always
> starts only first time. I start the IDE, compile and run the program (f.e.
> Hello.pas) and close IDE. But when I want to start IDE again it crashes
> with this error message:
>
> It occurs in both my systems - in FreeDOS and in Windows 98. To be able to
> start IDE again in FreeDOS I have to reboot the computer. Under W98 is
> enough to restart the DOS command prompt.

Ditto for me, as already mentioned. Also I don't know why. But just running (and exiting) INSTALL.EXE seems to reset / fix whatever bit that needs to be twiddled, thus the IDE can run again (for one instance). So that's a horribly kludgy but acceptable workaround.

> So, unfortunately 2.6.2 is a bad version for DOS. The unability to restart
> the IDE is the killer bug.

If that's the only problem, then we're doing pretty good!

> I recomend to stay with version 2.4.2 or, of course, with 1.0.10

Why not 2.4.4?

No, seriously, I doubt 1.0.10 is better. Sure, they had better DOS support in those days, but MarcoV will tell you a billion reasons why that version is deprecated (and not even mirrored anymore).

BTW, 2.6.2 "full install with sources" (FreeDOS, DOSLFN) chokes for me on DOCSRC.ZIP's options.tex for some odd reason. Deleting that file from the .ZIP seems to avoid the issue. Maybe a rare installer (unzip) bug.

> > 2)
> > Again occured weird problems with environment settings when shelling
> into
> > DOS. Under W98 it cuts some environment variables.
>
> The only difference I see here is TMP and TEMP missing. Maybe that's
> intentional for some reason, dunno.

I noted the damaged environment starts with the new string "D:\WINDOWS\COMMAND.COM" which is 22 bytes long. Comparing that to the undamaged environment, it appears that this is an size/alignment issue when "patching" the environment under W98 since the P of TEMP variable is infact still present but has the first 3 bytes of it overwritten. The environment uses nul terminated strings between strings. Comparing environments:

1234567890123456789012345
TMP=D:\WINDOWS\TEMP
TEMP=D:\WINDOWS\TEMP
vs
1234567890123456789012345
D:\WINDOWS\COMMAND.COM
P=D:\WINDOWS\TEMP

The string "TMP=D:\WINDOWS\TEMP" is 19 bytes in length, with 3 bytes missing off the next variable TEMP (TEM) = 22 in total. Or to write it another way which should help show exactly what I am talking about:
1234567890123456789012345
TMP=D:\WINDOWS\TEMP TEMP=D:\WINDOWS\TEMP
D:\WINDOWS\COMMAND.COM P=D:\WINDOWS\TEMP

I have further thoughts but am too tired to write something sensible now.

Sure, I can avoid bug 18133 in my code. But what about code by someone else? And what about code of Freepascal itself? F.e. the unit Graph is heavily influenced by this bug. Im not sure but the MEM statements maybe occur in RTL units and maybe in IDE sources or even other sources.

And we can't be sure it does not occur in other constructs than MEM. If the use of inline assembly always 100% safe.


The 2.7.1 seems to be fixed BUT.
In the 2.7.1 snapshots isn't included the IDE. I hope, I really hope that it doesn't mean that the 2.7.x line will be without it.

I know that FPC is implementing still wider and wider the unicode. I am also affraid it also could make a problems in DOS as it does not have any unicode layer. (however you can detect the current codepage and use various DOS to unicode tables). The unicode could also make problems in console applications made f.e. in FreeVision. The default VGA text framebuffer uses 1 byte for char and 1 byte for attribute. The framebuffer defined in VIDEO unit accept this format. The VIDEO framebuffer could be changed to another format and made conversions between VIDEO and VGA framebuffers while UpdateScreen procedures but anyway - it again breaks existing applications...

---
DOS-u-akbar!

> The 2.7.1 seems to be fixed BUT.
> In the 2.7.1 snapshots isn't included the IDE. I hope, I really hope that
> it doesn't mean that the 2.7.x line will be without it.

Doubt it, MarcoV is a big fan of the IDE. (BTW, I can never seem to get the snapshot from FTP, it's always incomplete / corrupted.)

> I know that FPC is implementing still wider and wider the unicode. I am
> also affraid it also could make a problems in DOS as it does not have any
> unicode layer.

It would be very silly to kill the DOS port due to Unicode, but stranger things have happened. Besides, with no maintainer for so many years, it's a miracle they still build for DOS at all.

> > Seems installer was updated (and now w/ srcs) but haven't tested whether
> it
> > fixes the FAT32 bug or not.

At least SysUtils is still wrong but DOS unit's DiskSize, DiskFree work.

> Still haven't tried installing "full" (with or without DOSLFN) yet.

Installing "full with sources" chokes on DOCSRC.ZIP's "source/docs/options.tex" (write error). Even if I try again, only installing DOCSRC.ZIP to a new target directory, it still chokes.

> > All I see are snapshots, e.g. last one (GO32V2, at least) is
> > on FPC FTP dated from January 17, 2013.
>
> Newest GO32V2 snapshot is partially truncated / incomplete (no CDS in ZIP),
> so I couldn't test.

fpc-2.7.1.go32v2.zip (March 27, 7.2 MB) is still corrupted! Oh well.

> Scott Moore's website is back online (thankfully). I was worried there for
> a while. (I really need to finally package up P5 for DJGPP on iBiblio,
> presumably without the other OS cruft.)

Done, see here. I've run the GNUmakefile [EDIT: only with GPC, obviously] on Windows and Linux as well, and the builds all pass the tests. (E-mail me if interested [doubt it].)

> Mirrored old LZMA 4.42 conversion to FPC (and my small patch) to iBiblio.
> Also mirrored some other stuff

And Trixter's Monotone.