On Thu, 2026-07-02 at 09:57 +0100, Andrew Stubbs wrote:
> Ping.
Thanks for the patch, sorry about the delay in reviewing this.
Testing this is usually a bit of a pain. You don't happen to have a
precanned way of doing this by any chance? :)
Given that I rarely touch the world of rtx, you're probably in a better
position to assess this patch than me. Mostly looks good...
>
> On 26/06/2026 16:07, Andrew Stubbs wrote:
> > Adds new features to the GDB Python script that prints GCC
> > internals.
> >
> > - Improved output for rtx_def types.
> > - Also enable printing for rtx, const_rtx, rtx_insn, etc.
> > - Add icode details for rtx_insn.
> > - Add register details for REG nodes.
> >
> > Examples:
> > <rtx_def 0x12345678 (mem)>
> > <rtx_def 0x12345678 (reg:DI 685 pseudo)>
> > <rtx_def 0x12345678 (reg:DI 1 r1)>
> > <rtx_def 0x12345678 (insn 160 movdi_symbol)>
> >
> > gcc/ChangeLog:
> >
> > * gdbhooks.py (Rtx): Decode more things.
> > (RtxPrinter): Improve output.
> > (build_pretty_printer): Match more rtx types.
> > ---
> > gcc/gdbhooks.py | 54 ++++++++++++++++++++++++++++++++++----------
> > -----
> > 1 file changed, 38 insertions(+), 16 deletions(-)
> >
> > diff --git a/gcc/gdbhooks.py b/gcc/gdbhooks.py
> > index 616fc605f62..74f358d51f2 100644
> > --- a/gcc/gdbhooks.py
> > +++ b/gcc/gdbhooks.py
> > @@ -423,6 +423,9 @@ class Rtx:
> > def GET_CODE(self):
> > return self.gdbval['code']
> >
> > + def XINT(self, n):
> > + return self.gdbval['u']['fld'][n]['rt_int']
> > +
> > def GET_RTX_LENGTH(code):
> > val_rtx_length = gdb.parse_and_eval('rtx_length')
> > return intptr(val_rtx_length[code])
> > @@ -435,30 +438,45 @@ def GET_RTX_FORMAT(code):
> > val_rtx_format = gdb.parse_and_eval('rtx_format')
> > return val_rtx_format[code].string()
> >
> > +def get_insn_name(icode):
> > + return gdb.parse_and_eval('insn_data')[icode]['name'].string()
> > +
> > +def reg_name(regno):
> > + val_reg_names = gdb.parse_and_eval('reg_names')
> > + try:
> > + return val_reg_names[regno].string()
> > + except:
> > + return "pseudo"
> > +
> > +def mode_name(mode):
> > + return gdb.parse_and_eval('mode_name')[mode].string()
> > +
> > class RtxPrinter:
> > def __init__(self, gdbval):
> > self.gdbval = gdbval
> > self.rtx = Rtx(gdbval)
> >
> > def to_string (self):
> > - """
> > - For now, a cheap kludge: invoke the inferior's print
> > - function to get a string to use the user, and return an
> > empty
> > - string for gdb
> > - """
> > - # We use print_inline_rtx to avoid a trailing newline
> > - gdb.execute('call print_inline_rtx (stderr, (const_rtx)
> > %s, 0)'
> > - % intptr(self.gdbval))
> > - return ''
...in that you're replacing the terrible kludge I wrote for this back
in 2013...
> > -
> > - # or by hand; based on gcc/print-rtl.c:print_rtx
> > result = ('<rtx_def 0x%x'
> > % (intptr(self.gdbval)))
> > code = self.rtx.GET_CODE()
> > - result += ' (%s' % GET_RTX_NAME(code)
> > - format_ = GET_RTX_FORMAT(code)
> > - for i in range(GET_RTX_LENGTH(code)):
> > - print(format_[i])
> > + name = GET_RTX_NAME(code)
> > + result += ' (%s' % name
> > + if name == 'insn' or name == 'call_insn' or name ==
> > 'jump_insn':
> > + result += ' %d' % self.gdbval['u2']['insn_uid']
> > + icode = self.rtx.XINT(5)
(might want a trailing comment here: # INSN_CODE)
> > + if icode != -1:
> > + result += ' %s' % get_insn_name(icode)
> > + else:
> > + result += ' <unrecognized>'
> > + elif name == 'reg':
> > + regno = self.gdbval['u']['reg']['regno']
> > + mode = mode_name(self.gdbval['mode'])
> > + result += ':%s %d %s' % (mode, regno, reg_name(regno))
> > + #else:
> > + #format_ = GET_RTX_FORMAT(code)
> > + #for i in range(GET_RTX_LENGTH(code)):
> > + # print(format_[i])
...though I see that you're commenting out my old implementation for
this case. How well does it work? Should we just remove it? Is any
of it worth saving? IIRC I tried it, had issues with it, and found the
kludge easier, but that was 13 years ago.
One other nit: please update the multiline string at the top of the
file, in particular the "RTL expressions use a kludge" part, which is
no longer true with your improvement.
OK with those nits addressed.
Hope this sounds sane; thanks again for the patch.
Dave
