294333 - NSS fails to build with cygwin on Windows ME (MSVC6 SP5)

Status: VERIFIED FIXED

(NSS :: Build, defect)

Description

[Jacek Piskozub]

For several weeks I am unable to build mozilla suite (CVS trunk) on Windows ME.
It builds OK with --disable-crypto. However with normal settings (including
--enable-crypto) the build crashes in /mozilla/security/nss.

I'll attach the full output of "make all" done in /mozilla/security/manager
folder. It crashes exactly the same as the full build (at least the last screen
is the same).

Comment 1

[Jacek Piskozub]

Attachment: The output of "make all" done on /mozilla/security/manager

Please take note that MS linker (cl.exe) gets a wrong option (Ld instead of LD)
in mozilla/security/nss/cmd/shlibsign. Also earlier I get a series of errors
including an "unknown processor architekture" in
mozilla/security/nss/lib/softoken and
mozilla_source/mozilla/security/nss/lib/nss plus a linker syntax error in
mozilla_source/mozilla/security/nss/lib/ssl

Comment 2

[Jacek Piskozub]

I'll add that the build machine is a Pentium 4. I build mozilla also on another
Windows ME box (Pentium 3) without problems (the settings are practically
identical, including the same versions of O/S, cygwin and MSVC). I'm not sure
what the difference may be (unless it is using different drives for building
mozilla (D: in case of the "bad" machine, C: on the "good" one).

Comment 3

[Wan-Teh Chang]

What's the output of "uname -a" on this machine?

Comment 4

[Jacek Piskozub]

I'll check it tomorrow morning (European time) when I'll have access to that
"bad" computer. Now, I can say the its "good" twin (well, almost twin) responds
with:
WIN95 [HOST_NAME_CHANGED] 4.90 73010104 xx I386

Comment 5

[Wan-Teh Chang]

Thanks.  From that output I concluded that on the "good twin"
you are using the uname command from Netscape wintools.

I suspect that on the "bad twin" you are using Cygwin's uname
command.  I suspect that "uname -s" probably says something
like CYGWIN_ME-x.xx.  If so, could you modify
mozilla/security/coreconf/arch.mk and add a section for
CYGWIN_ME-x.xx (replace x.xx with what "uname -s" returns)
that looks like the sections for CYGWIN_95-4.0 and CYGWIN_98-4.10?

Comment 6

[Jacek Piskozub]

Yes, you must be right. 

I recently changed the PATH on the bad machine, because the windows "sort"
command started crashing cygwin bash, by moving d:\cygwin\bin to the front. The
sort crash was solved and that is when I discovered this problem. I wrongly
assumed that it started in the time I could not build mozilla due to the other
problem.

Actually in doing this I tried to copy the PATH sequence from the "good" twin
but I did a mistake. I see now that the good one has 
PATH=C:\PROGRA~1\MICROS~2\COMMON\MSDEV98\BIN;C:\PROGRA~1\MICROS~2\VC98\BIN;C:\PROGRA~1\MICROS~2\COMMON\TOOLS\WIN95;C:\PROGRA~1\MICROS~2\COMMON\TOOLS;C:\WINDOWS\SYSTEM;D:\WINTOOLS\BIN;C:\CYGWIN\BIN;C:\CYGWIN\USR\X11R6\BIN;C:\WINDOWS;C:\WINDOWS\COMMAND;C:\PROGRA~1\NETWOR~1\PGP;
meaning that not only cygwin is before c:\windows but wintools are in front of
cygwin (I am sure I forgot to move wintools when moving cygwin).

Tomorrow, I will modify arch.mk as suggested (even as I know now how to solve
the problem by changing the PATH sequence).

A note to myself: the only PATH sequence which now works on Windows ME is:
MSVC; wintools; cygwin; windows\command; (the Microsoft sort.exe resides here)
while windows; and windows\system; may be practically anywhere in the sequence

Comment 7

[Jacek Piskozub]

BTW, the cygwin uname on the "good twin" responds with:

CYGWIN_ME-4.90 turquoise 1.5.16(0.128/4/2) 2005-04-25 20:26 i686 unknown unknown
Cygwin

so I'm going to add tomorrow the following section to arch.mk:

ifeq ($(OS_ARCH),CYGWIN_ME-4.90)
	OS_ARCH   = CYGWIN_NT-4.0
	OS_TARGET = WIN95
endif

Comment 8

[Wan-Teh Chang]

Attachment: Proposed patch

Give this patch a try, and let me know if it works
on Windows Me (with Cygwin's uname).

Comment 9

[Jacek Piskozub]

The patch works. A fresh CVS tree build of the suite went OK.

Is there a chance of checking in this extremely non-risky patch before 1.8b2?

Comment 10

[Wan-Teh Chang]

Comment on attachment 183780 [details] [diff] [review]
Proposed patch

Please review.	This patch allows building NSS on Windows Me.

Comment 11

[Nelson Bolyard (seldom reads bugmail)]

Comment on attachment 183780 [details] [diff] [review]
Proposed patch

I'm not a cygwin user, but considering the similarity of these new lines to the
ones preceeding them, it looks like it shold work.  Jacek has confirmed that it
works.	Thanks Jacek.

Comment 12

[Wan-Teh Chang]

Jacek, I will request 1.8b3 approval because I check in
NSS fixes for Mozilla in batches.

I am surprised that it is still possible to build Mozilla
on Windows 9x :-)

Nelson, I've checked this patch in on the NSS trunk for
NSS 3.10.1.

Checking in arch.mk;
/cvsroot/mozilla/security/coreconf/arch.mk,v  <--  arch.mk
new revision: 1.18; previous revision: 1.17
done

Comment 13

[Wan-Teh Chang]

Comment on attachment 183780 [details] [diff] [review]
Proposed patch

Requesting mozilla1.8b3 and aviary1.1a2 approvals.

This is a simple, low-risk makefile change to enable
building NSS on Windows Me.  The patch has been reviewed
by cls and Nelson Bolyard and verified by the bug reporter.

Comment 14

[Mike Shaver (:shaver emeritus)]

Comment on attachment 183780 [details] [diff] [review]
Proposed patch

a=shaver