61966 - Math.max(), Math.min(), Math.random() failing on Linux

Status: VERIFIED WORKSFORME

(Core :: JavaScript Engine, defect, P3)

Description

[Phil Schwartau]

Three math testcases are failing on my Linux box. Also one string testcase 
that relies on Math.max() and Math.min()

                      ecma/Math/15.8.2.11.js
                      ecma/Math/15.8.2.12.js
                      ecma/Math/15.8.2.14.js
                      ecma/String/15.5.4.7-2.js


ecma/Math/15.8.2.14.js is the worst: it crashes. But it's peculiar;
if I do this in the JS shell, it does not crash and the test passes:

./js
js> load('../../tests/ecma/shell.js');
js> load('../../tests/ecma/Math/15.8.2.14.js')
15.8.2.14 Math.random()
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
             .
            etc. 
            etc.


HOWEVER, the test driver does this, and I get a crash (on Linux):

./js -f ../../tests/ecma/shell.js -f ../../tests/ecma/Math/15.8.2.14.js
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
             .
            etc. 
            etc.
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Math.random() = pass PASSED!
Segmentation fault(core dumped)


Note this is happening to me only on Linux, not on WinNT. I tried to use gdb
to get a stack trace, but was unsuccessful at passing the -f arguments.


Will attach details from the test driver on all four testcases -

Comment 1

[Phil Schwartau]

Attachment: Linux test failure details

Comment 2

[Mike McCabe]

I'm seeing intermittent failures running the tests from the browser; each
clobber (with different patches) seems to give me a different result, all
involving failures with NaN and Infinity.  One promising note is that I saw two
tiny double values that differed in the last few places as different between two
tests.  We might be looking at uninitialized memory.

Comment 3

[Phil Schwartau]

Resolving WORKSFORME on Linux with JS shell built on 2001-02-22.
Also works on WinNT, Mac. 

Comment 4

[Phil Schwartau]

Marking Verified - 

Comment 5

[Phil Schwartau]

Reopening bug. These testcases are failing, and again only on Linux:
all rely on Math.max(), Math.min(), Math.random() in some way.

                      ecma/Math/15.8.2.11.js
                      ecma/Math/15.8.2.12.js
                      ecma/Math/15.8.2.14.js
                      ecma/String/15.5.4.7-1.js 
                      ecma/String/15.5.4.7-2.js

cc'ing Patrick, Brendan in case they are aware of any recent JS changes
that may have caused this- 

Comment 6

[Phil Schwartau]

Attachment: New New Linux test failure details

Comment 7

[Phil Schwartau]

If you compare the new test failures against the old, they differ.
For example: 
                                OLD                    NEW
ecma/Math/15.8.2.11.js       compare-errors           segfault        
ecma/Math/15.8.2.12.js       compare-errors           segfault

ecma/Math/15.8.2.14.js is failing because Math.random() == Infinity 
on my Linux box now. We expect Math.random() to lie between 0 and 1.

Comment 8

[Brendan Eich [:brendan]]

Phil, what stack do you get (segfault implies core file dropped, so hop into gdb
and bt)?  Did you clobber and rebuild after the fix for bug 8112-reopen went in
and was updated in your tree?

/be

Comment 9

[Phil Schwartau]

js> load('../../tests/js1_4/shell.js')
js> load('../../tests/ecma/Math/15.8.2.11.js')  
15.8.2.11 Math.max(x, y)

Program received signal SIGSEGV, Segmentation fault.
0x8097b37 in math_max (cx=0x80e4b78, obj=0x80e6ab0, argc=0,
argv=0x80ee3e0,                                                rval=0xbfffe520)
at jsmath.c:233
233         jsdouble x, z = *cx->runtime->jsNegativeInfinity;

(gdb) bt
#0  0x8097b37 in math_max (cx=0x80e4b78, obj=0x80e6ab0, argc=0,
argv=0x80ee3e0,                                            rval=0xbfffe520) at
jsmath.c:233
#1  0x807f148 in js_Invoke (cx=0x80e4b78, argc=0, flags=0) at jsinterp.c:813
#2  0x809117d in js_Interpret (cx=0x80e4b78, result=0xbfffe7a8)
at                                                                            
jsinterp.c:2706
#3  0x807f759 in js_Execute (cx=0x80e4b78, chain=0x80e6340,
script=0x80fae60,                       down=0x0, special=0, result=0xbfffe7a8)
at jsinterp.c:992
#4  0x8052b30 in JS_ExecuteScript (cx=0x80e4b78,
obj=0x80e6340,                                          script=0x80fae60,
rval=0xbfffe7a8) at jsapi.c:3183         
#5  0x8049e87 in Load (cx=0x80e4b78, obj=0x80e6340, argc=1,
argv=0x80ee304,                                                rval=0xbfffe864)
at js.c:637
#6  0x807f148 in js_Invoke (cx=0x80e4b78, argc=1, flags=0) at jsinterp.c:813
#7  0x809117d in js_Interpret (cx=0x80e4b78, result=0xbffffb00)
at                                                                            
jsinterp.c:2706
#8  0x807f759 in js_Execute (cx=0x80e4b78, chain=0x80e6340,
script=0x80ec868,                       down=0x0, special=0, result=0xbffffb00)
at jsinterp.c:992
#9  0x8052b30 in JS_ExecuteScript (cx=0x80e4b78,
obj=0x80e6340,                                          script=0x80ec868,
rval=0xbffffb00) at jsapi.c:3183
#10 0x8049632 in Process (cx=0x80e4b78, obj=0x80e6340, filename=0x0) at js.c:371
#11 0x8049af1 in ProcessArgs (cx=0x80e4b78, obj=0x80e6340,
argv=0xbffffbc8,                                                                
argc=0) at js.c:529
#12 0x804c069 in main (argc=0, argv=0xbffffbc8) at js.c:2096

Comment 10

[Phil Schwartau]

Since that turned out to be almost unreadable, let's try again! 

js> load('../../tests/js1_4/shell.js')
js> load('../../tests/ecma/Math/15.8.2.11.js')  
15.8.2.11 Math.max(x, y)

Program received signal SIGSEGV, Segmentation fault.
0x8097b37 in math_max (cx=0x80e4b78, obj=0x80e6ab0, argc=0, argv=0x80ee3e0, 
                                          rval=0xbfffe520) at jsmath.c:233
233         jsdouble x, z = *cx->runtime->jsNegativeInfinity;

(gdb) bt
#0  0x8097b37 in math_max (cx=0x80e4b78, obj=0x80e6ab0, argc=0, argv=0x80ee3e0, 
rval=0xbfffe520) at jsmath.c:233
#1  0x807f148 in js_Invoke (cx=0x80e4b78, argc=0, flags=0) at jsinterp.c:813
#2  0x809117d in js_Interpret (cx=0x80e4b78, result=0xbfffe7a8) at 
jsinterp.c:2706
#3  0x807f759 in js_Execute (cx=0x80e4b78, chain=0x80e6340, script=0x80fae60, 
down=0x0, special=0, result=0xbfffe7a8) at jsinterp.c:992
#4  0x8052b30 in JS_ExecuteScript (cx=0x80e4b78, obj=0x80e6340, 
script=0x80fae60, rval=0xbfffe7a8) at jsapi.c:3183
#5  0x8049e87 in Load (cx=0x80e4b78, obj=0x80e6340, argc=1, argv=0x80ee304, 
rval=0xbfffe864) at js.c:637
#6  0x807f148 in js_Invoke (cx=0x80e4b78, argc=1, flags=0) at jsinterp.c:813
#7  0x809117d in js_Interpret (cx=0x80e4b78, result=0xbffffb00) at 
jsinterp.c:2706
#8  0x807f759 in js_Execute (cx=0x80e4b78, chain=0x80e6340, script=0x80ec868, 
down=0x0, special=0, result=0xbffffb00) at jsinterp.c:992
#9  0x8052b30 in JS_ExecuteScript (cx=0x80e4b78, obj=0x80e6340, 
script=0x80ec868, rval=0xbffffb00) at jsapi.c:3183
#10 0x8049632 in Process (cx=0x80e4b78, obj=0x80e6340, filename=0x0) at js.c:371
#11 0x8049af1 in ProcessArgs (cx=0x80e4b78, obj=0x80e6340, argv=0xbffffbc8, 
argc=0) at js.c:529
#12 0x804c069 in main (argc=0, argv=0xbffffbc8) at js.c:2096

Comment 11

[Phil Schwartau]

Brendan's idea worked: 

>Did you clobber and rebuild after the fix for bug 8112-reopen went in
>and was updated in your tree?


Note I always do make -f Makefile.ref clean before I rebuild the shell.
This time I actually deleted the OBJ directories by hand, and for good measure, 
deleted the js/src/fdlibm directory and re-pulled it. Now, all the tests pass!

I'll resolve this once again as WORKSFORME. If anyone feels the 
stack trace reveals something deeper that is wrong, please reopen - 

Comment 12

[Phil Schwartau]

Note: On my Linux box, Math.random() once again lies between 0 and 1