AW: extension to -fdump-tree-*-raw

AW: extension to -fdump-tree-*-raw

[Ebke, Hans-Christian]

Hi Giovanni,
  I have to write this in Outlook, so I don't even try to get the quoting
right. Sorry. :-(

> -----Ursprüngliche Nachricht-----
> Von:	Giovanni Bajo [SMTP:rasky@develer.com]
> Gesendet am:	Freitag, 22. Juli 2005 12:27
> An:	Ebke, Hans-Christian
> Cc:	gcc@gcc.gnu.org
> Betreff:	Re: extension to -fdump-tree-*-raw
> 
> @54     string_cst       type: @61     strg: "wrong
> type:\n\0\0\xaf\x03\x03foo\"bar"  lngt: 19
> 
> This would not need a different special option.
> 
But it would break applications relying on the old format. Also it wouldn't
adress issues like this one:

@3      modify_expr      type: @5       op 0: @6       op 1: @7

Does "op" belong to the value of "type" or does it belong to the "0"? True,
this doesn't make parsing impossible but the parser needs to be really
specific to not fall into that trap. Having every "key-value pair" on a
separate line makes parsing much easier though it isn't quite as friendly to
the human reader anymore.

  Hans-Christian

Re: extension to -fdump-tree-*-raw

[Giovanni Bajo]

Ebke, Hans-Christian <h.ebke@kostal.com> wrote:

>   I have to write this in Outlook, so I don't even try to get the quoting
> right. Sorry. :-(

http://jump.to/outlook-quotefix

> But it would break applications relying on the old format.

There is no format either. dump-tree is *very* specific of GCC inners, and
it can dramatically changes between releases. OK, maybe not the syntax, but
the semantic. I wouldn't care of the syntax at that point.
-- 
Giovanni Bajo

Re: extension to -fdump-tree-*-raw

[Gabriel Dos Reis]

"Giovanni Bajo" <rasky@develer.com> writes:

| Ebke, Hans-Christian <h.ebke@kostal.com> wrote:
| 
| >   I have to write this in Outlook, so I don't even try to get the quoting
| > right. Sorry. :-(
| 
| http://jump.to/outlook-quotefix
| 
| > But it would break applications relying on the old format.
| 
| There is no format either. dump-tree is *very* specific of GCC inners, and
| it can dramatically changes between releases.

Yes, and I don't want to make any promise of any kind.

| OK, maybe not the syntax,

the syntax may change too.  It just did for 4.1.

-- Gaby

Re: extension to -fdump-tree-*-raw

[Ebke, Hans-Christian]

> -----Ursprüngliche Nachricht-----
> Von:	Gabriel Dos Reis [SMTP:gdr@integrable-solutions.net]
> 
> "Giovanni Bajo" <rasky@develer.com> writes:
> | Ebke, Hans-Christian <h.ebke@kostal.com> wrote:
> | 
> | >   I have to write this in Outlook, so I don't even try to get the
> quoting
> | > right. Sorry. :-(
> | 
> | http://jump.to/outlook-quotefix
> 
Outlook-Quotefix doesn't work with Outlook 97 (which I'm using). Thanks
anyway.

> | 
> | > But it would break applications relying on the old format.
> | 
> | There is no format either. dump-tree is *very* specific of GCC inners,
> and
> | it can dramatically changes between releases.
> 
> Yes, and I don't want to make any promise of any kind.
> 
> | OK, maybe not the syntax,
> 
> the syntax may change too.  It just did for 4.1.
> 
So if the syntax changed for 4.1 anyway and since my changes apply to the
4.0.1 sources I figure there's no need for my patches, right?

  Hans-Christian

Re: extension to -fdump-tree-*-raw

[Giovanni Bajo]

Ebke, Hans-Christian <h.ebke@kostal.com> wrote:

> So to resolve that problem I took the gcc 4.0.1 source code and patched
> tree.h and tree-dump.c. The patched version introduces two new options for
> -fdump-tree: The "parseable" option which produces unambiguous and easier
to
> parse but otherwise similar output to "raw" and the "maskstringcst" option
> which produces output with the string constants masked since this makes
> parsing the output even easier and I'm not interested in the string
> constants.


You could write some code to escape special characters, so to write
something like:

@54     string_cst       type: @61     strg: "wrong
type:\n\0\0\xaf\x03\x03foo\"bar"  lngt: 19

This would not need a different special option.
-- 
Giovanni Bajo

extension to -fdump-tree-*-raw

[Ebke, Hans-Christian]

Hi,

  wanting to obtain gcc's parse tree I found the -fdump-tree option.
Specifically the -fdump-tree-original-raw output contains all of the
information I need. When I tried to write a parser for the tree dump I
noticed, however that the output might contain ambiguities which can't be
resolved and render parsing the output correctly impossible.
For example, if your code contains a string constant with a colon you might
get into trouble:

warning("wrong type: foobar");

This code will make -fdump-tree-original-raw produce something like this:

@54     string_cst       type: @61     strg: wrong type: foobar  lngt: 19

Now try to make your parser extract the correct type.

Similar problem occur when a string constant contains "\n" or even "\0". The
former will produce a linebreak in the output making it hard if not
impossible to parse. The latter will produce output where the string
constant is truncated after the \0 character event though the "lngt"
property specifies the length of the whole, non-truncated string.

So to resolve that problem I took the gcc 4.0.1 source code and patched
tree.h and tree-dump.c. The patched version introduces two new options for
-fdump-tree: The "parseable" option which produces unambiguous and easier to
parse but otherwise similar output to "raw" and the "maskstringcst" option
which produces output with the string constants masked since this makes
parsing the output even easier and I'm not interested in the string
constants.

My question is: Does anybody think that these new features are of any use to
someone else but me and does anybody want to have a look at my patches and
maybe merge them with the official gcc sources?

  Hans-Christian