591152 - mozilla/js/src/configure: line 10134: test: : integer expression expected

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[ISHIKAWA, Chiaki]

Attachment: Patch to ./mozilla/js/src/configure.in

During the config/build cycle of TB3, I noticed a warning generated 
by ./mozilla/js/src/configure which looks very suspicious.

I reported the problem and analsys in the newsgroup posting.

http://groups.google.co.jp/group/mozilla.dev.builds/browse_thread/thread/c58a1b7825bf95a4#

I found that a code in configure looks like this at the line where the warning is generated:

0133  
10134   if test -z "$MACOS_DEPLOYMENT_TARGET" -o "$MACOS_DEPLOYMENT_TARGET" -ge "100300"; then
10135  
          ... blahblahblah ... 

The "-o" should be rewritten as "|| test " so that the null-string is not
passed to `test'.
Under my linux configuration, MACOS_DEPLOYMENT_TARGET is null.

Now I figured the problem is in configure.in and very easy to fix.
So I am attaching the patch to ./mozilla/ms/src/configure.in

TIA

Comment 1

[Tuukka Tolvanen (sp3000)]

checkin-needed?

Comment 2

[Tuukka Tolvanen (sp3000)]

oh, the other configure.in needs this too. (and comm) http://mxr.mozilla.org/comm-central/search?string=MACOS_DEPLOYMENT_TARGET

Comment 3

[Tuukka Tolvanen (sp3000)]

fwiw,
http://hg.mozilla.org/comm-central/rev/16853b562855
http://hg.mozilla.org/mozilla-central/rev/ee594fb0fa3e

Comment 4

[Tuukka Tolvanen (sp3000)]

there's also
    test -n "$XCODEBUILD_VERSION" -a "$XCODEBUILD_VERSION" -ge 620
which should be suspectible to the same sort of issue, but the other numerical comparison stuff seems ok

Comment 5

[ISHIKAWA, Chiaki]

Thank you for looking at the problem.

http://hg.mozilla.org/comm-central/rev/16853b562855
This one seems to need the fix.

http://hg.mozilla.org/mozilla-central/rev/ee594fb0fa3e
This one, also.

It is hard to tell which "test" requires changes because
we can't tell easily whether a particular string is defined at all.
In the case of "$MACOS_DEPLOYMENT_TARGET",
it seems to me
 - that certain version of MACOS deployment target may not
   define this string at all, and
 - if defined, there are version cases (larger than a given number)
   that need to be tested.

Depending on the strings chosen, they may or may not be defined
and being undefined means one thing to a particular test, and
so only the original coder, and the testers with various configurations
would be able to tell. (MACOS builder probably never see the
original problem I reported.)

The two patches I saw seemed to be an attempt to premature optimization
efforts. A series of extra exec()s  of "test" don't seem to
be a big issue to me for a configuration of big suite like TB3.
(Yes, it certainly adds time during configuration, but we have
bigger problems of real bugs and too many spurious compiler warnings that need
clean up before configure.in cleanup IMHO.)


TIA.

Comment 6

[Tuukka Tolvanen (sp3000)]

Attachment: central

covers gt/ge/lt/le/eq that were touched (ok, so there were only ge)

Comment 7

[Tuukka Tolvanen (sp3000)]

Attachment: comm

Comment 8

[Tuukka Tolvanen (sp3000)]

Attachment: central

oh bother. can't have blame for lines with bashisms ;>

Comment 9

[Tuukka Tolvanen (sp3000)]

Attachment: comm

Comment 10

[Justin Wood (:Callek)]

Comment on attachment 489980 [details] [diff] [review]
comm

I'm cheating and letting this get one single round of reviews, and will accept kyle's review here (provided the changed lines are == between m-c and c-c which they appear to be).

Comment 11

[Takanori MATSUURA]

Another choice:

if test -z "$MACOS_DEPLOYMENT_TARGET" -o "${MACOS_DEPLOYMENT_TARGET:-0}" -ge "100300"; then


Here's just for information.

Comment 12

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Comment on attachment 489978 [details] [diff] [review]
central

Maybe I still don't quite understand the problem here, my shell-fu is a bit weak, but why shouldn't

>     AC_MSG_CHECKING([for Unicode NSIS with major version == $REQ_NSIS_MAJOR_VER and minor version >= $MIN_NSIS_MINOR_VER])
>-    if test "$MAKENSISU_VER" == "" -o \
>-       ! "$MAKENSISU_MAJOR_VER" == "$REQ_NSIS_MAJOR_VER" -o \
>-       ! "$MAKENSISU_MINOR_VER" -ge $MIN_NSIS_MINOR_VER; then
>+    if test "$MAKENSISU_VER" = "" ||
>+       test ! "$MAKENSISU_MAJOR_VER" = "$REQ_NSIS_MAJOR_VER" -o \
>+            ! "$MAKENSISU_MINOR_VER" -ge $MIN_NSIS_MINOR_VER; then

be changed to '||' in both places?

Comment 13

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Other than that, these look fine, but I'd like to understand the question posted in comment 12 before I r+ these.

Comment 14

[Tuukka Tolvanen (sp3000)]

Nothing's checking MAKENSISU_MAJOR/MINOR_VER REQ_NSIS_MAJOR/MINOR_VER themselves for integerness or nonblankness there, and in a successful case it wants to look at both anyhow, so it should be no more broken written like that. The patch is just to isolate "not blank" tests from being part of the same syntax that will fail in the "blank" case.

Comment 15

[Tuukka Tolvanen (sp3000)]

> Nothing's checking MAKENSISU_MAJOR/MINOR_VER REQ_NSIS_MAJOR/MINOR_VER
> themselves for integerness or nonblankness there, and in a successful case it

("there" being the second "test" as patched -- if $MAKENSISU_VER is blank in the first test that does imply that the MAJOR/MINOR derived from it are as well, so the syntax would be broken if those two parts are not isolated.)

Comment 17

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Pushed the m-c parts in http://hg.mozilla.org/mozilla-central/rev/2fb4b186f3af

Leaving open until the c-c parts are landed.

Comment 18

[Mounir Lamouri (:mounir)]

Pushed in c-c:
https://hg.mozilla.org/comm-central/rev/319d4bf10b3d