Re: [PATCH] developer option: -fdump-generic-nodes; initial incorporation

[Richard Biener <richard.guenther@gmail.com>]

On Wed, Feb 28, 2024 at 9:25 AM Jakub Jelinek <jakub@redhat.com> wrote:
>
> On Wed, Feb 28, 2024 at 08:58:08AM +0100, Richard Biener wrote:
> > Incidentially this looks like something fit for a google summer of code project.
> > Ideally it would hook into print-tree.cc providing an alternate
> > structured output.
> > It currently prints in the style
> >
> >  <function_decl 0x7ffff71bc600 bswap16
> >     type <function_type 0x7ffff71ba5e8
> >         type <integer_type 0x7ffff702b540 short unsigned int public unsigned HI
> >             size <integer_cst 0x7ffff702d108 constant 16>
> >             unit-size <integer_cst 0x7ffff702d120 constant 2>
> >             align:16 warn_if_not_align:0 symtab:0 alias-set -1
> > canonical-type 0x7ffff702b540 precision:16 min <integer_cst
> > 0x7ffff702d138 0> max <integer_cst 0x7ffff702d0f0 65535>>
> >         QI
> >         size <integer_cst 0x7ffff702d048 constant 8>
> >         unit-size <integer_cst 0x7ffff702d060 constant 1>
> > ...
> >
> > where you can see it follows tree -> tree edges up to some depth
> > (and avoids repeated expansion).  When debugging that's all I have
> > and I have to follow edges by matching up the raw addresses printed,
> > re-dumping those that didn't get expanded.  HTML would be indeed
> > so much nicer here (and a more complete output).
>
> I think keeping the current format of what is printed but optionally just
> turn all those addresses in there into hyperlinks which would expand the
> other trees would be nice.  Maybe also allow just hovering on the link and
> show the other tree printed might be nice too.
> Folding it all into just <function_decl ... bswap <type link> ...>
> would mean one can't quickly access just the min/max or fn return type
> etc.  We might need some parameter how deep to go (and/or how many trees to
> dump at most) so that we don't dump into HTML gigabytes of data when asking
> to print say a large BIND_EXPR into HTML.

Btw, when in a debugging session I'd be fine with something like

(gdb) html-tree $3

creating a temporary file and spawning a window using the default .html
file handler with this being implemented in the python bindings similar
to how we handle dot-fn.  I realize that some terminals might support
parts of that "inline" (I bet my xterm doesn't ;)).

I think HTML might be machine-readable enough for this purpose
so indirecting via JSON or XML or whatnot isn't necessary (we don't
aim for this to be a way to extract GENERIC and feed it to another compiler).

Richard.

>
>         Jakub
>