Really crummy error messages, was Re: Ada files now checked in

Really crummy error messages, was Re: Ada files now checked in

[Phil Edwards]

On Sun, Oct 07, 2001 at 04:19:16PM +0200, Gabriel Dos Reis wrote:
> dewar@gnat.com writes:
> 
> | j.adb:3:04: "xyz" is undefined
> | 
> | compared to the c message
> | 
> | j.c:1: `asdf' undeclared (first use this function)
> 
> For a non-native English speaker like me, I have always questioned
> myself why the latter choice (which I find odd and difficult to read)
> is current practice and not the former.  FWIW, I would suggest that we
> switch to Robert's suggestion.

Even native English speakers, like me, get confused, and here's why:

For "first use this function" to be a correct English sentence [fragment],
there is only one possible parse, and it's not what is intended.  "First use
this function" is an /imperative/ with "use" as a verb (pronounced with a 'z'
sound), and always leaves new users wondering, "first use /which/ function?"

But GCC wants "use" to be a noun, pronounced with an 's' sound.  What we
really want to say is any of:

    j.c:1: `asdf' undeclared (first use in this function)
    j.c:1: `asdf' undeclared (first time used in this function)
    j.c:1: `asdf' undeclared (first occurrence in this function)

(Although maybe "is undeclared" would be better for C than "is undefined.")

Please, let's fix this.


Phil

-- 
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace.  We seek
not your counsel, nor your arms.  Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen.            - Samuel Adams

Re: Really crummy error messages

[Florian Weimer]

Phil Edwards <pedwards@disaster.jaj.com> writes:

> For "first use this function" to be a correct English sentence [fragment],
> there is only one possible parse, and it's not what is intended.  "First use
> this function" is an /imperative/ with "use" as a verb (pronounced with a 'z'
> sound), and always leaves new users wondering, "first use /which/ function?"

I think this error message is dating back to the old K&R
prototype-less days, when you declared functions implicitly by
calling ("using") them.  After a call such as "foo()", you can take
the address of the function by just writing "foo" (without further
declarations).

At least that's my guess.  In any case, the error message is really
misleading, close to "tagged type must be declared at library
level". ;-)

Re: Really crummy error messages, was Re: Ada files now checked in

[Zack Weinberg]

On Sun, Oct 07, 2001 at 05:35:47PM -0400, Phil Edwards wrote:
> 
>     j.c:1: `asdf' undeclared (first use in this function)
>     j.c:1: `asdf' undeclared (first time used in this function)
>     j.c:1: `asdf' undeclared (first occurrence in this function)
> 
> (Although maybe "is undeclared" would be better for C than "is undefined.")
> 
> Please, let's fix this.

Hmm...

$ cat >test.c
int foo() { baz; }
$ gcc test.c
test.c: In function `foo':
test.c:1: `baz' undeclared (first use in this function)
test.c:1: (Each undeclared identifier is reported only once
test.c:1: for each function it appears in.)
$

Where is this "first use this function" coming from?

While we're bitching, I am sick of seeing

test.c:1: (Each undeclared identifier is reported only once
test.c:1: for each function it appears in.)

I think the error message is perfectly clear without that.

zw

Re: Really crummy error messages, was Re: Ada files now checked in

[Phil Edwards]

On Mon, Oct 08, 2001 at 11:09:03AM -0700, Zack Weinberg wrote:
> $ cat >test.c
> int foo() { baz; }
> $ gcc test.c
> test.c: In function `foo':
> test.c:1: `baz' undeclared (first use in this function)
> test.c:1: (Each undeclared identifier is reported only once
> test.c:1: for each function it appears in.)
> $
> 
> Where is this "first use this function" coming from?

cp/lex.c, around line 1290.


> While we're bitching, I am sick of seeing
> 
> test.c:1: (Each undeclared identifier is reported only once
> test.c:1: for each function it appears in.)
> 
> I think the error message is perfectly clear without that.

I'd agree with this too.  Maybe GCC should have a beginner's mode?  :-)

Phil

-- 
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace.  We seek
not your counsel, nor your arms.  Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen.            - Samuel Adams

Re: Really crummy error messages, was Re: Ada files now checked in

[Tim Hollebeek]

On Mon, Oct 08, 2001 at 04:32:25PM -0400, Phil Edwards wrote:
> 
> > While we're bitching, I am sick of seeing
> > 
> > test.c:1: (Each undeclared identifier is reported only once
> > test.c:1: for each function it appears in.)
> > 
> > I think the error message is perfectly clear without that.
> 
> I'd agree with this too.  Maybe GCC should have a beginner's mode?  :-)

how about --verbose-errors, and have both a one-line and paragraph
explanation of each error?

  test.c:1: Message always goes here
  test.c:1: (iff --verbose errors, extended explanation
  test.c:1:  more...
  test.c:1:  etc)

This way you can normally use the normal options, and stick in
--verbose-errors when you see a new message or one you don't
understand.

-Tim

Re: Really crummy error messages, was Re: Ada files now checked in

[Fergus Henderson]

On 08-Oct-2001, Tim Hollebeek <tim@hollebeek.com> wrote:
> On Mon, Oct 08, 2001 at 04:32:25PM -0400, Phil Edwards wrote:
> > 
> > > While we're bitching, I am sick of seeing
> > > 
> > > test.c:1: (Each undeclared identifier is reported only once
> > > test.c:1: for each function it appears in.)
> > > 
> > > I think the error message is perfectly clear without that.
> > 
> > I'd agree with this too.  Maybe GCC should have a beginner's mode?  :-)
> 
> how about --verbose-errors, and have both a one-line and paragraph
> explanation of each error?
> 
>   test.c:1: Message always goes here
>   test.c:1: (iff --verbose errors, extended explanation
>   test.c:1:  more...
>   test.c:1:  etc)
> 
> This way you can normally use the normal options, and stick in
> --verbose-errors when you see a new message or one you don't
> understand.

We've used almost exactly that approach in the Mercury compiler,
and IMHO it works reasonably well.

One drawback with this approach as you have outlined it is that
beginners, who are the ones that most need the verbose error messages,
don't know to use the `--verbose-errors' option.  To alleviate this
problem, at the end of all the compilation errors, the Mercury compiler
outputs the following additional message:

	Recompile with `--verbose-errors' for more information.

-- 
Fergus Henderson <fjh@cs.mu.oz.au>  |  "I have always known that the pursuit
The University of Melbourne         |  of excellence is a lethal habit"
WWW: < http://www.cs.mu.oz.au/~fjh >  |     -- the last words of T. S. Garp.

Re: Really crummy error messages, was Re: Ada files now checked in

[Florian Weimer]

Fergus Henderson <fjh@cs.mu.oz.au> writes:

>> > > test.c:1: (Each undeclared identifier is reported only once
>> > > test.c:1: for each function it appears in.)

>> how about --verbose-errors, and have both a one-line and paragraph
>> explanation of each error?

> We've used almost exactly that approach in the Mercury compiler,
> and IMHO it works reasonably well.

The Ada compiler writes "(more references follow)" if the same
undeclared identifier is referenced multiple times.  Such an error
message is probably most helpful in this context, but I don't know if
you can implement it for the C compiler easily.

Re: Really crummy error messages, was Re: Ada files now checked in

[Phil Edwards]

On Wed, Oct 10, 2001 at 04:31:33PM +1000, Fergus Henderson wrote:
> One drawback with this approach as you have outlined it is that
> beginners, who are the ones that most need the verbose error messages,
> don't know to use the `--verbose-errors' option.  To alleviate this
> problem, at the end of all the compilation errors, the Mercury compiler
> outputs the following additional message:
> 
> 	Recompile with `--verbose-errors' for more information.

I'd probably get just as annoyed seeing that message repeatedly as I
would seeing the "(each doohickey is reported only once for each flurble)"
message that started this subthread.  :-)

This is one of those few areas where I feel the compiler should take
parameters from an environment variable.  Maybe leave "beginner mode"
messages on by default, but allow others of us to set

    export GCC_DIAGNOSTICS=nobeginner,spaced_out,ANSI_color_highlights

to turn off the extra messages, and also do indenting like

    http://gcc.gnu.org/ml/libstdc++/2000-q2/msg00318.html

with colors, etc.

Maybe only the driver would check for environment variables, and then insert
the appropriate -fdiagnostic options into the command line for the compiler.


Phil

-- 
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace.  We seek
not your counsel, nor your arms.  Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen.            - Samuel Adams

Re: Really crummy error messages, was Re: Ada files now checked in

[Zack Weinberg]

On Mon, Oct 08, 2001 at 04:32:25PM -0400, Phil Edwards wrote:
> On Mon, Oct 08, 2001 at 11:09:03AM -0700, Zack Weinberg wrote:
> > $ cat >test.c
> > int foo() { baz; }
> > $ gcc test.c
> > test.c: In function `foo':
> > test.c:1: `baz' undeclared (first use in this function)
> > test.c:1: (Each undeclared identifier is reported only once
> > test.c:1: for each function it appears in.)
> > $
> > 
> > Where is this "first use this function" coming from?
> 
> cp/lex.c, around line 1290.

Ah, the C++ compiler.  You didn't say which it was.

Hmm, there seems to be something else wrong:

$ cat test.c
int foo() { baz; }
int bar() { baz; }
$ g++ test.c
test.c: In function `int foo()':
test.c:1: `baz' undeclared (first use this function)
test.c:1: (Each undeclared identifier is reported only once for each function 
   it appears in.)

$ gcc test.c
test.c: In function `foo':
test.c:1: `baz' undeclared (first use in this function)
test.c:1: (Each undeclared identifier is reported only once
test.c:1: for each function it appears in.)
test.c: In function `bar':
test.c:2: `baz' undeclared (first use in this function)

They really ought to give identical error messages.

zw