Linker diagnostics for the incorrect options .

Linker diagnostics for the incorrect options .

[Umesh Kalappa]

Hi All ,

Linker error handling for non  options or incorrect options should be
robust for instance like
$gcc -c test.c
$ld -non-static test.o

The linker  interpretation for  "-n" as a standard option for  nmagic ,then
"o" as output filename  by name "n-static" which is the dynamic executable .

So linker should be robust here with proper diagnostic and we  would like
to fix the same,before that we would like to hear from the experts here.

  Thank you in advance .
~Umesh




<#m_-1933998605313114137_DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>

Re: Linker diagnostics for the incorrect options .

[Nick Clifton]

Hi Umesh,

> Linker error handling for non  options or incorrect options should be
> robust for instance like
> $gcc -c test.c
> $ld -non-static test.o
> 
> The linker  interpretation for  "-n" as a standard option for  nmagic ,then
> "o" as output filename  by name "n-static" which is the dynamic executable .

This is really a feature of the getopt functions provided by the libiberty
library.  Note that the correct behaviour does happen if you use two dashes
instead of one:

  % ld --non-static test.o
  ld: unrecognized option '--non-static'
  ld: use the --help option for usage information

Essentially this is a user education issue.  Long options should be proceeded 
by double dashes.  Single dashes may work, sometimes, but weird effects can
also occur.

Cheers
  Nick

Re: Linker diagnostics for the incorrect options .

[Fangrui Song]

On Tue, Sep 22, 2020 at 5:31 AM Nick Clifton via Binutils
<binutils@sourceware.org> wrote:
>
> Hi Umesh,
>
> > Linker error handling for non  options or incorrect options should be
> > robust for instance like
> > $gcc -c test.c
> > $ld -non-static test.o
> >
> > The linker  interpretation for  "-n" as a standard option for  nmagic ,then
> > "o" as output filename  by name "n-static" which is the dynamic executable .
>
> This is really a feature of the getopt functions provided by the libiberty
> library.  Note that the correct behaviour does happen if you use two dashes
> instead of one:
>
>   % ld --non-static test.o
>   ld: unrecognized option '--non-static'
>   ld: use the --help option for usage information
>
> Essentially this is a user education issue.  Long options should be proceeded
> by double dashes.  Single dashes may work, sometimes, but weird effects can
> also occur.
>
> Cheers
>   Nick
>

This is a conflict between the grouped short options syntax and single
dash prefixed long options.

% touch an ommand ':)'
% ld.bfd -you -can -ofcourse -use -this -Long -command -Line ':)'
:)

For newer long options, we should encourage EXACTLY_TWO_DASHES
(ld/lexsup.c). I have not seen grouped short options in a very large
code base and FreeBSD.
Maybe we can deprecate grouped short options syntax for ld.

Re: Linker diagnostics for the incorrect options .

[Nick Clifton]

[-- Attachment #1: Type: text/plain, Size: 818 bytes --]

Hi Fangrui,

> This is a conflict between the grouped short options syntax and single
> dash prefixed long options.
> 
> % touch an ommand ':)'
> % ld.bfd -you -can -ofcourse -use -this -Long -command -Line ':)'
> :)

he!

> For newer long options, we should encourage EXACTLY_TWO_DASHES

Definitely.

> (ld/lexsup.c). I have not seen grouped short options in a very large
> code base and FreeBSD.
> Maybe we can deprecate grouped short options syntax for ld.

I like this idea.  How about the attached patch ?  With it applied I
get output like this:

  % ld -vn -non-static -e 0 -v -e2 hello.o
  GNU ld (GNU Binutils) 2.35.50.20200923
  ld: Warning: grouped short command line options are deprecated: -vn
  ld: Error: unable to disambiguate: -non-static (did you mean --non-static ?)

  % echo $?
  1

Cheers
  Nick

[-- Attachment #2: ld-deprecate-grouped-options.patch --]
[-- Type: text/x-patch, Size: 1131 bytes --]

diff --git a/ld/lexsup.c b/ld/lexsup.c
index b8f066745f..f906508037 100644
--- a/ld/lexsup.c
+++ b/ld/lexsup.c
@@ -735,6 +735,21 @@ parse_args (unsigned argc, char **argv)
 	  optind = last_optind;
 	  optc = getopt_long (argc, argv, "-", really_longopts, &longind);
 	}
+      /* Attempt to detect grouped short options,  eg: "-non-start".
+	 Accepting such options is error prone as it is not clear if the user
+	 intended "-n -o n-start" or "--non-start".  */
+      else if (longind == 0  /* This is a short option.  */
+	       && optc > 32  /* It is a valid option.  */
+	       && optind != last_optind /* It is not a repeat of the previous option.  */
+	       && optc != argv[last_optind][1] /* The character is not the second character of argv[last_optind].  */
+	       )
+	{
+	  if (optarg)
+	    einfo (_("%F%P: Error: unable to disambiguate: %s (did you mean -%s ?)\n"),
+		   argv[last_optind], argv[last_optind]);
+	  else
+	    einfo (_("%P: Warning: grouped short command line options are deprecated: %s\n"), argv[last_optind]);
+	}
 
       if (ldemul_handle_option (optc))
 	continue;

Re: Linker diagnostics for the incorrect options .

[Fangrui Song]

On Wed, Sep 23, 2020 at 2:31 AM Nick Clifton <nickc@redhat.com> wrote:
>
> Hi Fangrui,
>
> > This is a conflict between the grouped short options syntax and single
> > dash prefixed long options.
> >
> > % touch an ommand ':)'
> > % ld.bfd -you -can -ofcourse -use -this -Long -command -Line ':)'
> > :)
>
> he!
>
> > For newer long options, we should encourage EXACTLY_TWO_DASHES
>
> Definitely.
>
> > (ld/lexsup.c). I have not seen grouped short options in a very large
> > code base and FreeBSD.
> > Maybe we can deprecate grouped short options syntax for ld.
>
> I like this idea.  How about the attached patch ?  With it applied I
> get output like this:
>
>   % ld -vn -non-static -e 0 -v -e2 hello.o
>   GNU ld (GNU Binutils) 2.35.50.20200923
>   ld: Warning: grouped short command line options are deprecated: -vn
>   ld: Error: unable to disambiguate: -non-static (did you mean --non-static ?)
>
>   % echo $?
>   1
>
> Cheers
>   Nick

Thanks! Perhaps    ` && optind != last_optind /* It is not a repeat of
the previous option.  */` can be deleted... Nobody repeats a short
option for ld,
unless they want to have fun with:

% Debug/ld/ld-new -vvvvvv
GNU ld (GNU Binutils) 2.35.50.20201002
GNU ld (GNU Binutils) 2.35.50.20201002
GNU ld (GNU Binutils) 2.35.50.20201002
GNU ld (GNU Binutils) 2.35.50.20201002
GNU ld (GNU Binutils) 2.35.50.20201002
GNU ld (GNU Binutils) 2.35.50.20201002

Re: Linker diagnostics for the incorrect options .

[Nick Clifton]

Hi Fangrui,

> Thanks! Perhaps    ` && optind != last_optind /* It is not a repeat of
> the previous option.  */` can be deleted... Nobody repeats a short
> option for ld,
> unless they want to have fun with:
> 
> % Debug/ld/ld-new -vvvvvv
> GNU ld (GNU Binutils) 2.35.50.20201002
> GNU ld (GNU Binutils) 2.35.50.20201002
> GNU ld (GNU Binutils) 2.35.50.20201002
> GNU ld (GNU Binutils) 2.35.50.20201002

Awww, you mean real users won't want to do that ? :-)

Advice accepted.  I have applied the patch with that line removed.

Cheers
  Nick

Re: Linker diagnostics for the incorrect options .

[Matthias Klose]

On 9/23/20 11:31 AM, Nick Clifton via Binutils wrote:
> Hi Fangrui,
> 
>> This is a conflict between the grouped short options syntax and single
>> dash prefixed long options.
>>
>> % touch an ommand ':)'
>> % ld.bfd -you -can -ofcourse -use -this -Long -command -Line ':)'
>> :)
> 
> he!
> 
>> For newer long options, we should encourage EXACTLY_TWO_DASHES
> 
> Definitely.
> 
>> (ld/lexsup.c). I have not seen grouped short options in a very large
>> code base and FreeBSD.
>> Maybe we can deprecate grouped short options syntax for ld.
> 
> I like this idea.  How about the attached patch ?  With it applied I
> get output like this:
> 
>   % ld -vn -non-static -e 0 -v -e2 hello.o
>   GNU ld (GNU Binutils) 2.35.50.20200923
>   ld: Warning: grouped short command line options are deprecated: -vn
>   ld: Error: unable to disambiguate: -non-static (did you mean --non-static ?)
> 
>   % echo $?
>   1

this causes PR ld/27050:

$ ld.bfd -no-pie
ld.bfd: Error: unable to disambiguate: -no-pie (did you mean --no-pie ?)