[rfc] Completion tests failing

[rfc] Completion tests failing

[Daniel Jacobowitz]

I don't like failing tests... especially ones that time out after absurd
periods of time.  So, I've been looking at the testsuite, particularly
completion.exp today.

The test failing for me right now is completion for 'p "a<tab>'.  The
comments, and the testsuite, say that this is a string literal and should
not be completed.  The code tries to complete it as a location, though, and
decides to run it through location_completer. make_symbol_completion_list
notices the stray ", and decides not to allow completion; but
make_source_files_completion_list does no such thing.

Which behavior would people prefer?  I can update the testsuite to match. 
I'd also like to add expressions like:

        -re "^p \"a\[^\"\]*\"$"     { fail "complete 'p \"a'" }

if it is not supposed to complete; that is, if gdb completes it to something
and supplies the trailing quote, fail now, instead of waiting for timeout
and failing the next dozen tests too.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [rfc] Completion tests failing

[Elena Zannoni]

Daniel Jacobowitz writes:
 > I don't like failing tests... especially ones that time out after absurd
 > periods of time.  So, I've been looking at the testsuite, particularly
 > completion.exp today.
 > 
 > The test failing for me right now is completion for 'p "a<tab>'.  The
 > comments, and the testsuite, say that this is a string literal and should
 > not be completed.  The code tries to complete it as a location, though, and
 > decides to run it through location_completer. make_symbol_completion_list
 > notices the stray ", and decides not to allow completion; but
 > make_source_files_completion_list does no such thing.
 > 
 > Which behavior would people prefer?  I can update the testsuite to match. 
 > I'd also like to add expressions like:
 > 
 >         -re "^p \"a\[^\"\]*\"$"     { fail "complete 'p \"a'" }
 > 
 > if it is not supposed to complete; that is, if gdb completes it to something
 > and supplies the trailing quote, fail now, instead of waiting for timeout
 > and failing the next dozen tests too.
 > 
 > -- 
 > Daniel Jacobowitz                           Carnegie Mellon University
 > MontaVista Software                         Debian GNU/Linux Developer


Daniel, is this related to the problem that Corinna was seeing?

http://sources.redhat.com/ml/gdb-patches/2001-08/msg00299.html

Elena

Re: [rfc] Completion tests failing

[Daniel Jacobowitz]

On Wed, Sep 26, 2001 at 09:34:06PM -0400, Elena Zannoni wrote:
> Daniel Jacobowitz writes:
>  > I don't like failing tests... especially ones that time out after absurd
>  > periods of time.  So, I've been looking at the testsuite, particularly
>  > completion.exp today.
>  > 
>  > The test failing for me right now is completion for 'p "a<tab>'.  The
>  > comments, and the testsuite, say that this is a string literal and should
>  > not be completed.  The code tries to complete it as a location, though, and
>  > decides to run it through location_completer. make_symbol_completion_list
>  > notices the stray ", and decides not to allow completion; but
>  > make_source_files_completion_list does no such thing.
>  > 
>  > Which behavior would people prefer?  I can update the testsuite to match. 
>  > I'd also like to add expressions like:
>  > 
>  >         -re "^p \"a\[^\"\]*\"$"     { fail "complete 'p \"a'" }
>  > 
>  > if it is not supposed to complete; that is, if gdb completes it to something
>  > and supplies the trailing quote, fail now, instead of waiting for timeout
>  > and failing the next dozen tests too.
>  > 
>  > -- 
>  > Daniel Jacobowitz                           Carnegie Mellon University
>  > MontaVista Software                         Debian GNU/Linux Developer
> 
> 
> Daniel, is this related to the problem that Corinna was seeing?
> 
> http://sources.redhat.com/ml/gdb-patches/2001-08/msg00299.html

Thanks - I'd missed that!  Yes, it's the same problem.  I think that
Corinna missed the comments describing the expected outcome, though -
rather than papering over the problem by making '"a' ambiguous, it
looks as if '"a' is not intended to complete at all.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [rfc] Completion tests failing

[Eli Zaretskii]

> Date: Wed, 26 Sep 2001 21:13:36 -0400
> From: Daniel Jacobowitz <drow@mvista.com>
> 
> The test failing for me right now is completion for 'p "a<tab>'.  The
> comments, and the testsuite, say that this is a string literal and should
> not be completed.

This is no longer true, so the test suite should be changed.

> The code tries to complete it as a location, though, and
> decides to run it through location_completer. make_symbol_completion_list
> notices the stray ", and decides not to allow completion; but
> make_source_files_completion_list does no such thing.

This is intended behavior, I think: source files can have quotes in
them.

> Which behavior would people prefer?  I can update the testsuite to match. 
> I'd also like to add expressions like:
> 
>         -re "^p \"a\[^\"\]*\"$"     { fail "complete 'p \"a'" }
> 
> if it is not supposed to complete; that is, if gdb completes it to something
> and supplies the trailing quote, fail now, instead of waiting for timeout
> and failing the next dozen tests too.

Given the recent changes to completion, I really don't see any sense
in testing this.  I'd suggest to delete the test.

Re: [rfc] Completion tests failing

[Fernando Nasser]

Eli Zaretskii wrote:
> 
> Given the recent changes to completion, I really don't see any sense
> in testing this.  I'd suggest to delete the test.

OK with me.  We could also change it to accept the "t" completion
noted by Corinna.

Daniel, are you sending in a patch?

Regards,
Fernando

-- 
Fernando Nasser
Red Hat Canada Ltd.                     E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9

Re: [rfc] Completion tests failing

[Daniel Jacobowitz]

On Thu, Sep 27, 2001 at 08:00:40AM -0400, Fernando Nasser wrote:
> Eli Zaretskii wrote:
> > 
> > Given the recent changes to completion, I really don't see any sense
> > in testing this.  I'd suggest to delete the test.
> 
> OK with me.  We could also change it to accept the "t" completion
> noted by Corinna.
> 
> Daniel, are you sending in a patch?

I will today.  Since the new behavior seems to be desired, I'll send in
a testsuite patch that fixes the possibility of a timeout and makes the
test less ambiguous.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [rfc] Completion tests failing

[Daniel Jacobowitz]

On Thu, Sep 27, 2001 at 08:00:40AM -0400, Fernando Nasser wrote:
> Eli Zaretskii wrote:
> > 
> > Given the recent changes to completion, I really don't see any sense
> > in testing this.  I'd suggest to delete the test.
> 
> OK with me.  We could also change it to accept the "t" completion
> noted by Corinna.
> 
> Daniel, are you sending in a patch?

Actually, before I send in a patch, I have a question.  I've found
another completion bug.

I wanted to get a name that I knew would complete unambiguously, so I
tried using '"break.', which ought to complete to '"break.c"'.  It
completes instead to '"break.c ' - trailing space, no trailing quote.

Am I right about what ought to happen?  If so, I'll make this two
testcases, one failing, and then try to fix it.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [rfc] Completion tests failing

[Fernando Nasser]

Daniel Jacobowitz wrote:
> 
> On Thu, Sep 27, 2001 at 08:00:40AM -0400, Fernando Nasser wrote:
> > Eli Zaretskii wrote:
> > >
> > > Given the recent changes to completion, I really don't see any sense
> > > in testing this.  I'd suggest to delete the test.
> >
> > OK with me.  We could also change it to accept the "t" completion
> > noted by Corinna.
> >
> > Daniel, are you sending in a patch?
> 
> Actually, before I send in a patch, I have a question.  I've found
> another completion bug.
> 
> I wanted to get a name that I knew would complete unambiguously, so I
> tried using '"break.', which ought to complete to '"break.c"'.  It
> completes instead to '"break.c ' - trailing space, no trailing quote.
> 
> Am I right about what ought to happen?  If so, I'll make this two
> testcases, one failing, and then try to fix it.
> 

I am not 100% sure of what was Eli's intention, but I guess we would 
need to add a test to add the closing quote if the filename started 
with quote.  (The space at the end seems normal as the completer has
found a "complete" completion (i.e., a full file name) already).

Eli is in another time zone, so we may have to wait until tomorrow
to hear from him.

Thanks for finding/fixing this.

Regards,
Fernando

-- 
Fernando Nasser
Red Hat Canada Ltd.                     E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9

Re: [rfc] Completion tests failing

[Eli Zaretskii]

> Date: Thu, 27 Sep 2001 11:30:01 -0400
> From: Fernando Nasser <fnasser@redhat.com>
> 
> I am not 100% sure of what was Eli's intention, but I guess we would 
> need to add a test to add the closing quote if the filename started 
> with quote.

The code is already there, but the first completion attempt doesn't
``know'' that it completes a file name.  We discussed this nuisance in
the past.

> Eli is in another time zone, so we may have to wait until tomorrow
> to hear from him.

Well, I'm in another time zone allright, but not so far from you: I'm
only 6 hours ahead ;-)

Andrew, on the other hand...

Re: [rfc] Completion tests failing

[Daniel Jacobowitz]

On Thu, Sep 27, 2001 at 11:30:01AM -0400, Fernando Nasser wrote:
> I am not 100% sure of what was Eli's intention, but I guess we would 
> need to add a test to add the closing quote if the filename started 
> with quote.  (The space at the end seems normal as the completer has
> found a "complete" completion (i.e., a full file name) already).
> 
> Eli is in another time zone, so we may have to wait until tomorrow
> to hear from him.
> 
> Thanks for finding/fixing this.

OK, it seems that we do not add the closing quote if there was a .,
because of our readline interactions; thanks for explaining that, Eli.
I've added a test for it anyway, marked XFAIL.

OK to commit the attached?

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer


2001-09-27  Daniel Jacobowitz  <drow@mvista.com>

	* gdb.base/completion.exp: Remove incorrect 'p "a' test.
	Add tests for 'p "break' (pass) and 'p "break.' (xfail).

Index: completion.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.base/completion.exp,v
retrieving revision 1.6
diff -u -r1.6 completion.exp
--- completion.exp	2001/08/06 22:08:12	1.6
+++ completion.exp	2001/09/27 17:19:36
@@ -38,7 +38,9 @@
 #   "info ajksdlfk " no completions
 #   "info" " "
 #   "info " ambiguous (all info commands)
-#   "p \"a" no completions (string constant)
+#   "p \"break" unambiguous (completes to filename "break.c")
+#   "p \"break." unambiguous (should complete to "break.c" but does not,
+#	due to readline limitations)
 #   "p 'a" ambiguous (all symbols starting with a)
 #   "p b-a" ambiguous (all symbols starting with a)
 #   "p b-" ambiguous (all symbols)
@@ -349,20 +351,61 @@
         }
 
 
-send_gdb "p \"a\t"
+send_gdb "p \"break\t"
 sleep 1
 gdb_expect  {
-        -re "^p \"a\\\x07$"\
+        -re "^p \"break\\\x07$"\
             { send_gdb "\n"
               gdb_expect {
-                      -re "Unterminated string in expression\\..*$gdb_prompt $"\
-                                        { pass "complete 'p a'"}
-                      -re ".*$gdb_prompt $" { fail "complete 'p a'"}
-                      timeout           {fail "(timeout) complete 'p a'"}
+                      -re ".*$gdb_prompt $" { fail "complete 'p \"break'"}
+                      timeout           {fail "(timeout) complete 'p \"break'"}
                      }
             }
-        -re ".*$gdb_prompt $"       { fail "complete 'p \"a'" }
-        timeout         { fail "(timeout) complete 'p \"a'" }
+	-re "^p \"break\\.c\"$"\
+	    {	send_gdb "\n"
+		gdb_expect {
+		    -re ".*$gdb_prompt $" { pass "complete 'p \"break'"}
+		    timeout           {fail "(timeout) complete 'p \"break'"}
+		}
+	    }
+	-re "^p \"break.*$"
+	    {	send_gdb "\n"
+		gdb_expect {
+		    -re ".*$gdb_prompt $" { fail "complete 'p \"break'"}
+		    timeout           {fail "(timeout) complete 'p \"break'"}
+		}
+	    }
+        -re ".*$gdb_prompt $"       { fail "complete 'p \"break'" }
+        timeout         { fail "(timeout) complete 'p \"break'" }
+        }
+
+setup_xfail "*-*-*"
+send_gdb "p \"break.\t"
+sleep 1
+gdb_expect  {
+        -re "^p \"break\\.\\\x07$"\
+            { send_gdb "\n"
+              gdb_expect {
+                      -re ".*$gdb_prompt $" { fail "complete 'p \"break.'"}
+                      timeout           {fail "(timeout) complete 'p \"break.'"}
+                     }
+            }
+	-re "^p \"break\\.c\"$"\
+	    {	send_gdb "\n"
+		gdb_expect {
+		    -re ".*$gdb_prompt $" { pass "complete 'p \"break.'"}
+		    timeout           {fail "(timeout) complete 'p \"break.'"}
+		}
+	    }
+	-re "^p \"break\\..*$"
+	    {	send_gdb "\n"
+		gdb_expect {
+		    -re ".*$gdb_prompt $" { fail "complete 'p \"break.'"}
+		    timeout           {fail "(timeout) complete 'p \"break.'"}
+		}
+	    }
+        -re ".*$gdb_prompt $"       { fail "complete 'p \"break.'" }
+        timeout         { fail "(timeout) complete 'p \"break.'" }
         }
 
 send_gdb "p 'a\t"

Re: [rfc] Completion tests failing

[Eli Zaretskii]

> Date: Thu, 27 Sep 2001 13:23:02 -0400
> From: Daniel Jacobowitz <drow@mvista.com>
> 
> OK, it seems that we do not add the closing quote if there was a .,
> because of our readline interactions; thanks for explaining that, Eli.
> I've added a test for it anyway, marked XFAIL.
> 
> OK to commit the attached?

Fine with me, if Fernando doesn't object.

Re: [rfc] Completion tests failing

[Fernando Nasser]

Eli Zaretskii wrote:
> 
> > Date: Thu, 27 Sep 2001 13:23:02 -0400
> > From: Daniel Jacobowitz <drow@mvista.com>
> >
> > OK, it seems that we do not add the closing quote if there was a .,
> > because of our readline interactions; thanks for explaining that, Eli.
> > I've added a test for it anyway, marked XFAIL.
> >
> > OK to commit the attached?
> 
> Fine with me, if Fernando doesn't object.

I as just waiting for your opinion.

Yes, please check it in.

And thanks for the patch.

-- 
Fernando Nasser
Red Hat Canada Ltd.                     E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9

Re: [rfc] Completion tests failing

[Daniel Jacobowitz]

On Thu, Sep 27, 2001 at 04:44:02PM -0400, Fernando Nasser wrote:
> Eli Zaretskii wrote:
> > 
> > > Date: Thu, 27 Sep 2001 13:23:02 -0400
> > > From: Daniel Jacobowitz <drow@mvista.com>
> > >
> > > OK, it seems that we do not add the closing quote if there was a .,
> > > because of our readline interactions; thanks for explaining that, Eli.
> > > I've added a test for it anyway, marked XFAIL.
> > >
> > > OK to commit the attached?
> > 
> > Fine with me, if Fernando doesn't object.
> 
> I as just waiting for your opinion.
> 
> Yes, please check it in.
> 
> And thanks for the patch.

Checked in.  One more failure down the drain!

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [rfc] Completion tests failing

[Eli Zaretskii]

> Date: Thu, 27 Sep 2001 10:34:01 -0400
> From: Daniel Jacobowitz <drow@mvista.com>
> 
> I wanted to get a name that I knew would complete unambiguously, so I
> tried using '"break.', which ought to complete to '"break.c"'.  It
> completes instead to '"break.c ' - trailing space, no trailing quote.

Try this twice, one after the other, and you will see that only the
first time fails.  The second and all the successive times will
complete to `"break.c"'.

This is a known limitation of the current completion machinery: GDB
needs to set the correct word-delimiting characters, but Readline
doesn't give us a chance to do that before it breaks the user input
into words.  Thus, the first time our completion function is called,
the input is broken into words according to wrong delimiters.

We cannot fix this without some additional hooks provided by Readline.

Re: [rfc] Completion tests failing

[Daniel Jacobowitz]

On Thu, Sep 27, 2001 at 05:37:10PM +0200, Eli Zaretskii wrote:
> > Date: Thu, 27 Sep 2001 10:34:01 -0400
> > From: Daniel Jacobowitz <drow@mvista.com>
> > 
> > I wanted to get a name that I knew would complete unambiguously, so I
> > tried using '"break.', which ought to complete to '"break.c"'.  It
> > completes instead to '"break.c ' - trailing space, no trailing quote.
> 
> Try this twice, one after the other, and you will see that only the
> first time fails.  The second and all the successive times will
> complete to `"break.c"'.

What do you mean by twice?  I can't get `"break.' to complete to `"break.c"'
at all, no matter what I try.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [rfc] Completion tests failing

[Eli Zaretskii]

> Date: Thu, 27 Sep 2001 13:00:26 -0400
> From: Daniel Jacobowitz <drow@mvista.com>
> 
> On Thu, Sep 27, 2001 at 05:37:10PM +0200, Eli Zaretskii wrote:
> > > Date: Thu, 27 Sep 2001 10:34:01 -0400
> > > From: Daniel Jacobowitz <drow@mvista.com>
> > > 
> > > I wanted to get a name that I knew would complete unambiguously, so I
> > > tried using '"break.', which ought to complete to '"break.c"'.  It
> > > completes instead to '"break.c ' - trailing space, no trailing quote.
> > 
> > Try this twice, one after the other, and you will see that only the
> > first time fails.  The second and all the successive times will
> > complete to `"break.c"'.
> 
> What do you mean by twice?  I can't get `"break.' to complete to `"break.c"'
> at all, no matter what I try.

Sorry, I didn't realize the importance of starting with "break.,
including the period.  Anything else correctly completes to "break.c".

The reason for this is that Readline only passes our completion
function the part after the period, so we never see the "break part.

Re: [rfc] Completion tests failing

[Daniel Jacobowitz]

On Thu, Sep 27, 2001 at 09:08:14PM +0200, Eli Zaretskii wrote:
> > Date: Thu, 27 Sep 2001 13:00:26 -0400
> > From: Daniel Jacobowitz <drow@mvista.com>
> > 
> > On Thu, Sep 27, 2001 at 05:37:10PM +0200, Eli Zaretskii wrote:
> > > > Date: Thu, 27 Sep 2001 10:34:01 -0400
> > > > From: Daniel Jacobowitz <drow@mvista.com>
> > > > 
> > > > I wanted to get a name that I knew would complete unambiguously, so I
> > > > tried using '"break.', which ought to complete to '"break.c"'.  It
> > > > completes instead to '"break.c ' - trailing space, no trailing quote.
> > > 
> > > Try this twice, one after the other, and you will see that only the
> > > first time fails.  The second and all the successive times will
> > > complete to `"break.c"'.
> > 
> > What do you mean by twice?  I can't get `"break.' to complete to `"break.c"'
> > at all, no matter what I try.
> 
> Sorry, I didn't realize the importance of starting with "break.,
> including the period.  Anything else correctly completes to "break.c".

(on the first try, best I can see)

> The reason for this is that Readline only passes our completion
> function the part after the period, so we never see the "break part.

Well, that's not right either.  It completes unambiguously to break.c.
Oh, on the other hand I see that text contains the whole string and
word contains only \0.  This is definitely a guts-of-readline problem,
as readline thinks that the "." is a word separator, which is not
precisely correct.  Oh, well.  XFAIL for now, and we can poke at this
some more later.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer