errors and warnings lost when redirecting output to files in a shell

[GoogleCodeExporter]

How to find a warning:

donny@donny-desktop:~/src/hg/gpsee$ gsr -c 'warning=42'
JS strict warning #156 in command_line at line 1 - assignment to undeclared
variable warning

How to lose a warning:
donny@donny-desktop:~/src/hg/gpsee$ gsr -c 'warning=42' &> log
donny@donny-desktop:~/src/hg/gpsee$ cat log

I was giving nickg some code to run to help evaluate a bug report, but then
I discovered the bug I'm reporting here now. The code below demonstrates my
proposed use case, and the reason I think the current behavior is a bug:

RESULTFILE=bugreport.txt
(hg id                           &&
make clean                       &&
make build                       &&
make install                     &&
echo @testing /tmp/foo           ;
gsr -c 'require("/tmp/foo")'     ;
echo @testing /tmp/foo.js        ;
gsr -c 'require("/tmp/foo.js")'
)                                &> $RESULTFILE

Also -d doesn't help:

donny@donny-desktop:~/src/hg/gpsee$ gsr -c 'require("/foo/js")' &> f
donny@donny-desktop:~/src/hg/gpsee$ cat f
donny@donny-desktop:~/src/hg/gpsee$ gsr -dc 'require("/foo/js")' &> f
donny@donny-desktop:~/src/hg/gpsee$ cat f
donny@donny-desktop:~/src/hg/gpsee$ gsr -ddc 'require("/foo/js")' &> f
donny@donny-desktop:~/src/hg/gpsee$ cat f
donny@donny-desktop:~/src/hg/gpsee$ gsr -dddc 'require("/foo/js")' &> f
donny@donny-desktop:~/src/hg/gpsee$ cat f

I will take a look at this and try to ensure that -d will output the same
error and warning output even if the stderr is not a tty.

Wes please let me know if my intentions are pure and good.

Original issue reported on code.google.com by donny.vi...@gmail.com on 19 Apr 2010 at 2:01

[GoogleCodeExporter]

The basic problem is that:

1. -c and preload processing is independent of -d levels
2. Warnings and errors are reported on stderr
3. -c & preload sets a non-zero debug level only when stderr is not a TTY 
(assumption
is that you are running from cron or something)
4. You are redirecting stderr 

Generally speaking, we don't want errors and warnings going to non-TTY device 
unless
explicitly requested with a raised debug level -- however, you are using -c 
which is
debug-level agnostic.

So, we have two cases: 
 - people want -c to have a different-than-script debug level so that -c gives error
output when used alone
 - people want -c to have the same-as-script debug level so that -d causes -c to put
output out stderr

Frankly, my immediate thought is to remove -c entirely, because we keep wasting 
time
on it when it and it is confusing people. Confusing, conflicting features and
spaghetti code generally don't make for good software.

Original comment by wes%page...@gtempaccount.com on 19 Apr 2010 at 12:38

[GoogleCodeExporter]

I've had some more thoughts about this code, hopefully I'm a little clearer now 
that
that migraine is gone.

We need to decide if -c is going to get the same error handling as the main 
program
or not.  If so, we need to refactor gsr.c to get processFlags() happening 
before the
-c code is executed. This may mean an extra getopt loop.

We also need to consider the case where users truly want debug output on their
non-tty outputs.  What do you think about a -T flag which means "treat stderr 
and
stdout as tty, whether they are or not?"

Original comment by wes%page...@gtempaccount.com on 23 Apr 2010 at 2:13

[GoogleCodeExporter]

Original comment by wes@page.ca on 5 Sep 2012 at 3:24