Hello,
there is a discrepancy amongst compilers on how the default output streams behave in some situations. Consider the following test case:
(format *standard-output* "This goes to standard output.~%") (format *error-output* "This goes to error output.~%") (format *query-io* "This goes to query io.~%")
Now, launch you favorite Lisp on a terminal and load this file at startup in 3 different ways:
myfavoritelisp --the-load-option test.lisp myfavoritelisp --the-load-option test.lisp > log myfavoritelisp --the-load-option test.lisp > log 2>&1
SBCL and CMUCL behave in a way which I think is right: case #1: everything is printed on the terminal case #2: stdout goes in the file case #3: stdout and stderr go in the file (*query-io* stays on the tty)
ECL redirects the output of *query-io* in the file for cases #2 and #3. The behavior of stdout and stderr is the same as in SBCL and CMUCL.
Finally, LW, ACL, CLISP, CCL and ABCL redirect everything in the file in cases #2 and #3.
I'm ready to fill in bug reports for the last 6 compilers, but are these really bugs? I'm convinced that it is for stderr, but it may be arguable for *query-io*.
WDYT?
On May 22, 2012, at 10:58 , Didier Verna wrote:
Hello,
there is a discrepancy amongst compilers on how the default output streams behave in some situations. Consider the following test case:
(format *standard-output* "This goes to standard output.~%") (format *error-output* "This goes to error output.~%") (format *query-io* "This goes to query io.~%")
Now, launch you favorite Lisp on a terminal and load this file at startup in 3 different ways:
myfavoritelisp --the-load-option test.lisp myfavoritelisp --the-load-option test.lisp > log myfavoritelisp --the-load-option test.lisp > log 2>&1
SBCL and CMUCL behave in a way which I think is right: case #1: everything is printed on the terminal case #2: stdout goes in the file case #3: stdout and stderr go in the file (*query-io* stays on the tty)
ECL redirects the output of *query-io* in the file for cases #2 and #3. The behavior of stdout and stderr is the same as in SBCL and CMUCL.
Finally, LW, ACL, CLISP, CCL and ABCL redirect everything in the file in cases #2 and #3.
I'm ready to fill in bug reports for the last 6 compilers, but are these really bugs? I'm convinced that it is for stderr, but it may be arguable for *query-io*.
WDYT?
I think this is the usual story. Behavior should be agreed upon, once the "platform" is given.
MA
-- Marco Antoniotti, Associate Professor tel. +39 - 02 64 48 79 01 DISCo, Università Milano Bicocca U14 2043 http://bimib.disco.unimib.it Viale Sarca 336 I-20126 Milan (MI) ITALY
Please note that I am not checking my Spam-box anymore. Please do not forward this email without asking me first.
Didier Verna didier@lrde.epita.fr writes:
Hello,
there is a discrepancy amongst compilers on how the default output streams
s/compilers/implementations/
behave in some situations. Consider the following test case:
(format *standard-output* "This goes to standard output.~%") (format *error-output* "This goes to error output.~%") (format *query-io* "This goes to query io.~%")
Now, launch you favorite Lisp on a terminal and load this file at startup in 3 different ways:
myfavoritelisp --the-load-option test.lisp myfavoritelisp --the-load-option test.lisp > log myfavoritelisp --the-load-option test.lisp > log 2>&1
SBCL and CMUCL behave in a way which I think is right: case #1: everything is printed on the terminal case #2: stdout goes in the file case #3: stdout and stderr go in the file (*query-io* stays on the tty)
I agree.
ECL redirects the output of *query-io* in the file for cases #2 and #3. The behavior of stdout and stderr is the same as in SBCL and CMUCL.
Finally, LW, ACL, CLISP, CCL and ABCL redirect everything in the file in cases #2 and #3.
I'm ready to fill in bug reports for the last 6 compilers, but are these really bugs? I'm convinced that it is for stderr, but it may be arguable for *query-io*.
WDYT?
Those are not bugs. Those are unspecified, implementation dependent behaviors. Well *query-io* is somewhat informally specified, so we definitely could consider that it should use /dev/tty on systems providing it.
AFAIK, /dev/tty is not POSIX (posix doesn't specify /dev/). So this specification would have to be restricted to systems providing /dev/tty or a similar feature.
You should write a CDR, and then indeed ask vendors to implement it.
On Tue, 22 May 2012 10:58:27 +0200, Didier Verna said:
Hello,
there is a discrepancy amongst compilers on how the default output streams behave in some situations. Consider the following test case:
(format *standard-output* "This goes to standard output.~%") (format *error-output* "This goes to error output.~%") (format *query-io* "This goes to query io.~%")
Now, launch you favorite Lisp on a terminal and load this file at startup in 3 different ways:
myfavoritelisp --the-load-option test.lisp myfavoritelisp --the-load-option test.lisp > log myfavoritelisp --the-load-option test.lisp > log 2>&1
SBCL and CMUCL behave in a way which I think is right: case #1: everything is printed on the terminal case #2: stdout goes in the file case #3: stdout and stderr go in the file (*query-io* stays on the tty)
ECL redirects the output of *query-io* in the file for cases #2 and #3. The behavior of stdout and stderr is the same as in SBCL and CMUCL.
Finally, LW, ACL, CLISP, CCL and ABCL redirect everything in the file in cases #2 and #3.
I'm ready to fill in bug reports for the last 6 compilers, but are these really bugs? I'm convinced that it is for stderr, but it may be arguable for *query-io*.
WDYT?
It's a feature request, not a bug (the ANSI standard even says that it is common for the initial values of *error-output* and *standard-output* be the same stream).
IMHO, having two independently-buffered streams connected to the same underlying destination in the interactive case #1 is a recipe for confusion.
FWIW, for CMUCL whole story looks like this (from memory, so take this with a grain of salt) -- SBCL does the same, unsurprisingly.
There are streams *STDIN*, *STDOUT*, *STDERR*, and *TTY*. If /dev/tty cannot be opened, *TTY* is simply
(make-two-way-stream *stdin* *stdout*)
*STANDARD-INPUT*, *STANDARD-OUTPUT*, and *ERROR-OUTPUT* start out as synonym streams for *STDIN*, *STDOUT*, *STDERR*.
*TERMINAL-IO*, *QUERY-IO* and *DEBUG-IO* start out as synonym streams for *TTY*.
Neither two-way-streams nor synonym streams get buffering of their own: anything written to them is immediately written to the underlying stream.
It's not a perfect setup, but it's not half bad, IMO.
Cheers,
-- nikodemus
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hello,
As others have pointed out, this does not appear to be a bug with respect to the hyperspec.
But in my opinion, those implementations which send *error-output* to the same place as *standard-output* on POSIX platforms do have a bug. There is a quite reasonable expectation for how stdout and stderr should behave on those platforms, and if a language implementation does not conform to that behavior, it seems like it can be considered a bug.
Conformance with the hyperspec isn't the only standard for considering if some behavior is a bug, is it?
- -Jason
On 05/22/2012 10:58 AM, Didier Verna wrote:
Hello,
there is a discrepancy amongst compilers on how the default output streams behave in some situations. Consider the following test case:
(format *standard-output* "This goes to standard output.~%") (format *error-output* "This goes to error output.~%") (format *query-io* "This goes to query io.~%")
Now, launch you favorite Lisp on a terminal and load this file at startup in 3 different ways:
myfavoritelisp --the-load-option test.lisp myfavoritelisp --the-load-option test.lisp > log myfavoritelisp --the-load-option test.lisp > log 2>&1
SBCL and CMUCL behave in a way which I think is right: case #1: everything is printed on the terminal case #2: stdout goes in the file case #3: stdout and stderr go in the file (*query-io* stays on the tty)
ECL redirects the output of *query-io* in the file for cases #2 and #3. The behavior of stdout and stderr is the same as in SBCL and CMUCL.
Finally, LW, ACL, CLISP, CCL and ABCL redirect everything in the file in cases #2 and #3.
I'm ready to fill in bug reports for the last 6 compilers, but are these really bugs? I'm convinced that it is for stderr, but it may be arguable for *query-io*.
WDYT?
"Jason S. Cornez" jcornez@alum.mit.edu writes:
Conformance with the hyperspec isn't the only standard for considering if some behavior is a bug, is it?
The principle of least surprise should be considered whenever possible, at the least.
-
Didier Verna -
Ian Dalton -
Jason S. Cornez -
Marco Antoniotti -
Martin Simmons -
Nikodemus Siivola -
Pascal J. Bourguignon