improve syntax errors
Hello!
I have used GCC for decades and would like to contribute a little. :-)
I would like to see if I can improve the syntax errors.
Here is one example code:
int x = 3) + 0;
I have created this ugly experimental patch:
--- a/gcc/c/c-parser.c
+++ b/gcc/c/c-parser.c
@@ -2228,7 +2228,10 @@ c_parser_declaration_or_fndef (c_parser
*parser, bool fndef_ok,
}
else
{
- c_parser_error (parser, "expected %<,%> or %<;%>");
+ if (c_parser_next_token_is (parser, CPP_CLOSE_PAREN))
+ c_parser_error (parser, "extraneous )");
+ else
+ c_parser_error (parser, "expected %<,%> or %<;%>");
c_parser_skip_to_end_of_block_or_statement (parser);
return;
}
Before my patch:
test1.c:3:12: error: expected ‘,’ or ‘;’ before ‘)’ token
After my patch:
test1.c:3:12: error: extraneous ) before ‘)’ token
That is not perfect neither.
I have 2 questions..
* can somebody give me a hint how I improve the error message so it
does not say "before ) token"?
* how do I run the tests?
Basically I want that when there is a unmatched extra ) or } or ] then
it should just say "extraneous .." instead of "expected ',' or ';'.
Adding a ',' or ';' in the example code will not fix the syntax error.
Best regards,
Daniel Marjamäki
Re: improve syntax errors
On Thu, 2019-01-03 at 15:59 +0100, Daniel Marjamäki wrote:
> Hello!
>
> I have used GCC for decades and would like to contribute a little. :-
> )
Hi, and welcome!
> I would like to see if I can improve the syntax errors.
>
> Here is one example code:
>
> int x = 3) + 0;
>
> I have created this ugly experimental patch:
>
> --- a/gcc/c/c-parser.c
> +++ b/gcc/c/c-parser.c
> @@ -2228,7 +2228,10 @@ c_parser_declaration_or_fndef (c_parser
> *parser, bool fndef_ok,
> }
> else
> {
> - c_parser_error (parser, "expected %<,%> or %<;%>");
> + if (c_parser_next_token_is (parser, CPP_CLOSE_PAREN))
> + c_parser_error (parser, "extraneous )");
> + else
> + c_parser_error (parser, "expected %<,%> or %<;%>");
> c_parser_skip_to_end_of_block_or_statement (parser);
> return;
> }
>
> Before my patch:
> test1.c:3:12: error: expected ‘,’ or ‘;’ before ‘)’ token
>
> After my patch:
> test1.c:3:12: error: extraneous ) before ‘)’ token
> That is not perfect neither.
Thanks for trying it out.
A minor nit: the ")" is a source code construct and thus should be
quoted in the message, by wrapping it in %< and %> like in the existing
code:
c_parser_error (parser, "extraneous %<)%>");
or by using %qs to quote a const const *:
c_parser_error (parser, "extraneous %qs", ")");
(FWIW, the word "extraneous" seems a bit "jargony" to me, how about
"stray %qs token"? (giving: "error: stray ')' token") I'm not sure
though)
> I have 2 questions..
> * can somebody give me a hint how I improve the error message so it
> does not say "before ) token"?
The before ')' token is being supplied due to the use of
c_parser_error, which calls c_parse_error, which adds a
"before SOMETHING"
suffix to the message before calling into the diagnostic subsystem.
You could use:
error_at (token->location, "some message");
to go directly to the underlying diagnostic API to avoid getting the
"before SOMETHING" suffix.
> * how do I run the tests?
You might want to look at this guide I put together:
https://dmalcolm.fedorapeople.org/gcc/newbies-guide/index.html
which among other things has some notes on running tests (and writing
new ones), and on stepping through gcc in a debugger.
> Basically I want that when there is a unmatched extra ) or } or ]
> then
> it should just say "extraneous .." instead of "expected ',' or ';'.
> Adding a ',' or ';' in the example code will not fix the syntax
> error.
I wonder how much we want to special-case this. Are you thinking about
the case where there's a stray symbol in the code (perhaps due to a
stray keypress, or unfinished manual edits)? Perhaps we could have a
predicate for determining if a token can never make sense in the given
context, and have something like:
if (c_parser_next_token_is_meaningless_p (parser))
complain_about_stray_token (parser);
else
...
or somesuch (we use a "_p" suffix for predicates).
It might make sense to emit a fix-it hint suggesting the removal of the
stray token.
Much of these ideas could probably apply to the C++ frontend as well
(annoyingly, not much of this code is shared between C and C++).
> Best regards,
> Daniel Marjamäki
Hope this is helpful, and welcome again.
Dave
Re: improve syntax errors
On Thu, 3 Jan 2019 at 15:40, David Malcolm wrote:
>
> On Thu, 2019-01-03 at 15:59 +0100, Daniel Marjamäki wrote:
> > Hello!
> >
> > I have used GCC for decades and would like to contribute a little. :-
> > )
>
> Hi, and welcome!
>
> > I would like to see if I can improve the syntax errors.
> >
> > Here is one example code:
> >
> > int x = 3) + 0;
> >
> > I have created this ugly experimental patch:
> >
> > --- a/gcc/c/c-parser.c
> > +++ b/gcc/c/c-parser.c
> > @@ -2228,7 +2228,10 @@ c_parser_declaration_or_fndef (c_parser
> > *parser, bool fndef_ok,
> > }
> > else
> > {
> > - c_parser_error (parser, "expected %<,%> or %<;%>");
> > + if (c_parser_next_token_is (parser, CPP_CLOSE_PAREN))
> > + c_parser_error (parser, "extraneous )");
> > + else
> > + c_parser_error (parser, "expected %<,%> or %<;%>");
> > c_parser_skip_to_end_of_block_or_statement (parser);
> > return;
> > }
> >
> > Before my patch:
> > test1.c:3:12: error: expected ‘,’ or ‘;’ before ‘)’ token
> >
> > After my patch:
> > test1.c:3:12: error: extraneous ) before ‘)’ token
>
> > That is not perfect neither.
>
> Thanks for trying it out.
>
> A minor nit: the ")" is a source code construct and thus should be
> quoted in the message, by wrapping it in %< and %> like in the existing
> code:
>
> c_parser_error (parser, "extraneous %<)%>");
>
> or by using %qs to quote a const const *:
>
> c_parser_error (parser, "extraneous %qs", ")");
>
>
> (FWIW, the word "extraneous" seems a bit "jargony" to me, how about
> "stray %qs token"? (giving: "error: stray ')' token") I'm not sure
> though)
Maybe "unmatched"?
> > I have 2 questions..
> > * can somebody give me a hint how I improve the error message so it
> > does not say "before ) token"?
>
> The before ')' token is being supplied due to the use of
> c_parser_error, which calls c_parse_error, which adds a
> "before SOMETHING"
> suffix to the message before calling into the diagnostic subsystem.
> You could use:
>
> error_at (token->location, "some message");
>
> to go directly to the underlying diagnostic API to avoid getting the
> "before SOMETHING" suffix.
>
> > * how do I run the tests?
>
> You might want to look at this guide I put together:
> https://dmalcolm.fedorapeople.org/gcc/newbies-guide/index.html
> which among other things has some notes on running tests (and writing
> new ones), and on stepping through gcc in a debugger.
>
> > Basically I want that when there is a unmatched extra ) or } or ]
> > then
> > it should just say "extraneous .." instead of "expected ',' or ';'.
> > Adding a ',' or ';' in the example code will not fix the syntax
> > error.
>
> I wonder how much we want to special-case this. Are you thinking about
> the case where there's a stray symbol in the code (perhaps due to a
> stray keypress, or unfinished manual edits)? Perhaps we could have a
> predicate for determining if a token can never make sense in the given
> context, and have something like:
>
> if (c_parser_next_token_is_meaningless_p (parser))
> complain_about_stray_token (parser);
> else
> ...
>
> or somesuch (we use a "_p" suffix for predicates).
Right, why emit a special diagnostic for 3) but not 3]?
Re: syntax errors
Thank you for the quick reply.
> how about "stray %qs token"?
I will change.
> I wonder how much we want to special-case this. Are you thinking about
> the case where there's a stray symbol in the code (perhaps due to a
> stray keypress, or unfinished manual edits)?
At the moment I only think about ) } ]
I would like to focus on only those 3 to start with. but it sounds
good to prepare for more stray tokens later.
> if (c_parser_next_token_is_meaningless_p (parser))
> complain_about_stray_token (parser);
> else
> ...
sure!
> It might make sense to emit a fix-it hint suggesting the removal of the
> stray token.
It is 50% chance that the closing paranthesis should be removed. Maybe
there is a missing "(".
Maybe the error message should indicate that.. something like "either
there is missing "(" or this ")" is a stray token".
Best regards,
Daniel Marjamäki
Mais praticidade para suas compras online.
http://tr.santander.com.br/f/a/PPmjU4rVj0zTRnyGYvUq3A~~/AQA~/RgReEMorP0Q4aHR0cDovL3d3dy53My5vcmcvVFIveGh0bWwxL0RURC94aHRtbDEtdHJhbnNpdGlvbmFsLmR0ZCJXBXVuZWFyQgoAAitFLlxZy7umUgtnY2NAZ251Lm9yZ1gEAA~~> http://tr.santander.com.br/f/a/bMOvSvHWQ1mS1duk35pSMw~~/AQA~/RgReEMorP0QdaHR0cDovL3d3dy53My5vcmcvMTk5OS94aHRtbCJXBXVuZWFyQgoAAitFLlxZy7umUgtnY2NAZ251Lm9yZ1gEAA~~> Mais praticidade para suas compras online. Mais praticidade para suas compras online. http://tr.santander.com.br/f/a/qOPX2TC57dysBXs2ymna7A~~/AQA~/RgReEMorP0Q4aHR0cHM6Ly93d3cuc2FudGFuZGVyLmNvbS5ici9uZXdzbGV0dGVyLzEwNjM3OGIvaW1nLmpwZyJXBXVuZWFyQgoAAitFLlxZy7umUgtnY2NAZ251Lm9yZ1gEAA~~ width="117" height="72" border="0" style="display:block; border:0;" /> ROSA, http://tr.santander.com.br/f/a/cMMr3-h53Cb6kyDNSapR6g~~/AQA~/RgReEMorP0Q6aHR0cHM6Ly93d3cuc2FudGFuZGVyLmNvbS5ici9uZXdzbGV0dGVyLzEwNjM3OGIvaW1nMDIuanBnIlcFdW5lYXJCCgACK0UuXFnLu6ZSC2djY0BnbnUub3JnWAQA width="596" height="168" border="0" style="display:block; border:0;" /> http://tr.santander.com.br/f/a/FcovhG5MZd9yYxkJ4VUOMA~~/AQA~/RgReEMorP0Q6aHR0cHM6Ly93d3cuc2FudGFuZGVyLmNvbS5ici9uZXdzbGV0dGVyLzEwNjM3OGIvaW1nMDMuanBnIlcFdW5lYXJCCgACK0UuXFnLu6ZSC2djY0BnbnUub3JnWAQA alt="" width="596" height="307" style="display:block; border:0;" border="0" /> http://tr.santander.com.br/f/a/7TOYWXg2ORAZ_qegKdhgcA~~/AQA~/RgReEMorP0Q4aHR0cHM6Ly93d3cuc2FudGFuZGVyLmNvbS5ici9uZXdzbGV0dGVyLzEwNjM3OGIvdHh0LmpwZyJXBXVuZWFyQgoAAitFLlxZy7umUgtnY2NAZ251Lm9yZ1gEAA~~ alt="" width="596" height="280" style="display:block; border:0;" border="0" /> http://tr.santander.com.br/f/a/1VlJSmg6wql_-lvuDklHoA~~/AQA~/RgReEMorP0Q6aHR0cHM6Ly93d3cuc2FudGFuZGVyLmNvbS5ici9uZXdzbGV0dGVyLzEwNjM3OGIvaW1nMDQuanBnIlcFdW5lYXJCCgACK0UuXFnLu6ZSC2djY0BnbnUub3JnWAQA alt="" width="596" height="100" style="display:block; border:0;" border="0" /> http://tr.santander.com.br/f/a/VZlkt9NOyFHlKVkCEmHG1w~~/AQA~/RgReEMorP0Q6aHR0cHM6Ly93d3cuc2FudGFuZGVyLmNvbS5ici9uZXdzbGV0dGVyLzEwNjM3OGIvaW1nMDUuanBnIlcFdW5lYXJCCgACK0UuXFnLu6ZSC2djY0BnbnUub3JnWAQA width="596" height="317" border="0" style="display:block; border:0;" /> 19530898 http://tr.santander.com.br/f/a/b6PM0HXEU1d30u_PHUrRKQ~~/AQA~/RgReEMorP0Q6aHR0cHM6Ly93d3cuc2FudGFuZGVyLmNvbS5ici9uZXdzbGV0dGVyLzEwNjM3OGIvaW1nMDYuanBnIlcFdW5lYXJCCgACK0UuXFnLu6ZSC2djY0BnbnUub3JnWAQA alt="" width="596" height="100" style="display:block; border:0;" border="0" /> http://tr.santander.com.br/f/a/tgwtFNzhUPNo4BtXf6fv0g~~/AQA~/RgReEMorP0R6aHR0cHM6Ly93ZWJzZXJ2aWNlLnVjYW1wYWlnbi51bmVhci5uZXQvVW1haWxUcmFja2luZy90LmFzcHg_cD03NzYyMjMwNTkmYz1NVEEwTkRjMU1RPT0mdXA9NTcxNzA3NzAyJmU9Z2NjQGdudS5vcmcmbD1NREU9JiJXBXVuZWFyQgoAAitFLlxZy7umUgtnY2NAZ251Lm9yZ1gEAA~~ />
Re: syntax errors
On Thu, 3 Jan 2019 at 17:03, Daniel Marjamäki wrote: > > Thank you for the quick reply. > > > how about "stray %qs token"? > > I will change. > > > I wonder how much we want to special-case this. Are you thinking about > > the case where there's a stray symbol in the code (perhaps due to a > > stray keypress, or unfinished manual edits)? > > At the moment I only think about ) } ] > > I would like to focus on only those 3 to start with. but it sounds > good to prepare for more stray tokens later. It doesn't scale to write a custom diagnostic for every invalid input character in every context.
Mtrace with no environment MALLOC_TRACE feature request mtrace_setpath.
Request for mtrace_setpath
In debugging and profiling PAM (Pluggable Authentication Modules) modules for
memory leaks in linux, mtrace cannot access environment variables as these do
not exist until the authentication completes (e.g. /etc/environment . The
MALLOC_TRACE path can only be supplied as an environment variable. Please add a
feature to allow the passing of a trace path to mtrace. The module workaround
is to add putenv("MALLOC_TRACE=/var/log/mypammemleak.log");
Globally changing the linux environment values to trace a security module
should not be an answer.
Please consider adding a mtrace_setpath(const char* path);
Note : pam_env.so creates environment variables afterward. I've checked
everything other than rebuilding PAM.
Module order does not matter.
Joe
Re: Mtrace with no environment MALLOC_TRACE feature request mtrace_setpath.
On Thu, 3 Jan 2019 at 21:18, Joseph Howard wrote: > > Request for mtrace_setpath I think you should have sent this to glibc, which is a completely separate project to GCC. See https://www.gnu.org/software/libc/libc.html
gcc-7-20190103 is now available
Snapshot gcc-7-20190103 is now available on ftp://gcc.gnu.org/pub/gcc/snapshots/7-20190103/ and on various mirrors, see http://gcc.gnu.org/mirrors.html for details. This snapshot has been generated from the GCC 7 SVN branch with the following options: svn://gcc.gnu.org/svn/gcc/branches/gcc-7-branch revision 267565 You'll find: gcc-7-20190103.tar.xzComplete GCC SHA256=9f8043c1ba5923ebbbf0a5ae1ab854aceed26232ed19a16920b554d283928327 SHA1=c2776c7a476936c348a52fbe8172c15407c7cad8 Diffs from 7-20181227 are available in the diffs/ subdirectory. When a particular snapshot is ready for public consumption the LATEST-7 link is updated and a message is sent to the gcc list. Please do not use a snapshot before it has been announced that way.
Re: improve syntax errors
On 1/3/19 8:40 AM, David Malcolm wrote:
On Thu, 2019-01-03 at 15:59 +0100, Daniel Marjamäki wrote:
Hello!
I have used GCC for decades and would like to contribute a little. :-
)
Hi, and welcome!
I would like to see if I can improve the syntax errors.
Here is one example code:
int x = 3) + 0;
I have created this ugly experimental patch:
--- a/gcc/c/c-parser.c
+++ b/gcc/c/c-parser.c
@@ -2228,7 +2228,10 @@ c_parser_declaration_or_fndef (c_parser
*parser, bool fndef_ok,
}
else
{
- c_parser_error (parser, "expected %<,%> or %<;%>");
+ if (c_parser_next_token_is (parser, CPP_CLOSE_PAREN))
+ c_parser_error (parser, "extraneous )");
+ else
+ c_parser_error (parser, "expected %<,%> or %<;%>");
c_parser_skip_to_end_of_block_or_statement (parser);
return;
}
Before my patch:
test1.c:3:12: error: expected ‘,’ or ‘;’ before ‘)’ token
After my patch:
test1.c:3:12: error: extraneous ) before ‘)’ token
That is not perfect neither.
Thanks for trying it out.
A minor nit: the ")" is a source code construct and thus should be
quoted in the message, by wrapping it in %< and %> like in the existing
code:
c_parser_error (parser, "extraneous %<)%>");
or by using %qs to quote a const const *:
c_parser_error (parser, "extraneous %qs", ")");
I would suggest the quoted form since it means less work for message
translantors. (Both forms are used in GCC messages but I think it
would be worth adopting the quoted form.)
(FWIW, the word "extraneous" seems a bit "jargony" to me, how about
"stray %qs token"? (giving: "error: stray ')' token") I'm not sure
though)
A helpful rule of thumb I like to use is checking which form is
prevalent in gcc/po/gcc.pot. In this case, there's just one
message with the word "extraneous" in all of GCC and 7 that
use "stray." There are also three messages that use the word
"unmatched", and for variety, five that use "unterminated" (not
all of them are interchangeable).
Martin
I have 2 questions..
* can somebody give me a hint how I improve the error message so it
does not say "before ) token"?
The before ')' token is being supplied due to the use of
c_parser_error, which calls c_parse_error, which adds a
"before SOMETHING"
suffix to the message before calling into the diagnostic subsystem.
You could use:
error_at (token->location, "some message");
to go directly to the underlying diagnostic API to avoid getting the
"before SOMETHING" suffix.
* how do I run the tests?
You might want to look at this guide I put together:
https://dmalcolm.fedorapeople.org/gcc/newbies-guide/index.html
which among other things has some notes on running tests (and writing
new ones), and on stepping through gcc in a debugger.
Basically I want that when there is a unmatched extra ) or } or ]
then
it should just say "extraneous .." instead of "expected ',' or ';'.
Adding a ',' or ';' in the example code will not fix the syntax
error.
I wonder how much we want to special-case this. Are you thinking about
the case where there's a stray symbol in the code (perhaps due to a
stray keypress, or unfinished manual edits)? Perhaps we could have a
predicate for determining if a token can never make sense in the given
context, and have something like:
if (c_parser_next_token_is_meaningless_p (parser))
complain_about_stray_token (parser);
else
...
or somesuch (we use a "_p" suffix for predicates).
It might make sense to emit a fix-it hint suggesting the removal of the
stray token.
Much of these ideas could probably apply to the C++ frontend as well
(annoyingly, not much of this code is shared between C and C++).
Best regards,
Daniel Marjamäki
Hope this is helpful, and welcome again.
Dave
