Fix stderr to output string with no decoration (fix #2063) #2751
Conversation
itchyny
force-pushed
the
fix-stderr-raw-string
branch
4 times, most recently
from
16cd5e7
to
10e6a61
Compare
itchyny
force-pushed
the
fix-stderr-raw-string
branch
from
10e6a61
to
d030ead
Compare
| @@ -232,6 +232,16 @@ grep "Expected string key after '{', not '\\['" $d/err > /dev/null | |||
| echo '{"x":"y",["a","b"]}' | $JQ --stream > /dev/null 2> $d/err || true | |||
| grep "Expected string key after ',' in object, not '\\['" $d/err > /dev/null | |||
|
|
|||
| # debug, stderr | |||
| $VALGRIND $Q $JQ -n '"test", {} | debug, stderr' >/dev/null | |||
There was a problem hiding this comment.
Should output be compared somehow here?
There was a problem hiding this comment.
Unrelated thought: I wonder if i would be nicer if we replaced $VALGRIND $Q $JQ by having jq in path be a wrapper script that does exec valgrind -q /path/to/real/jq/ ...? or do we don't want to not run via valgrind in some cases when it's enabled?
There was a problem hiding this comment.
I put this line just for valgrind (and I actually found a leak in the first commit), and I heard Nico said we don't care about the output of debug/0 somewhere. I strongly hope that JQ takes care of valgrind automatically but there might some difficulties in testing the standard error output.
There was a problem hiding this comment.
Aha valgrind can messes up stdout/err? by "hope that JQ takes care" do you mean it would be nice if $JQ pointed to some path that runs jq with valgrind?
Yes think i read that comment from @nicowilliams also, but even if we make no guarantees about stableness maybe it can make sense to test debug output so that we would notice if it changes by accident? maybe we already do somewhere?
There was a problem hiding this comment.
debug/0 should do something with its output. We could have a command-line option to send that to some file, or /dev/null/NUL, and debug/0 can decorate its output or not (but it helps that it does).
I don't really care about the mechanics of how valgrind or other memory debuggers get invoked, as long as a) such debuggers can be optional, b) they can be turned off (valgrind slows dev cycles). If the current thing makes it easy to forget to use valgrind, sure, let's fix it.
There was a problem hiding this comment.
Also, if in --run-tests we set a stderr callback, we could have syntax in .test files for checking stderr's output, though I don't think we need that given that a test here in shtest is just as well, and we only need stderr to test it.
|
Thanks! |
As written in the manual, the
stderrfilter should output values in raw and compact mode with no additional decoration. This PR fixes #2063.