Re: [PATCHv4] gdb/python/mi: create MI commands using python

[Tom Tromey <tom@tromey.com>]

>>>>> "Andrew" == Andrew Burgess via Gdb-patches <gdb-patches@sourceware.org> writes:

Andrew> This v4 patch is identical to the version I posted here:
Andrew>    https://sourceware.org/pipermail/gdb-patches/2022-February/185821.html

Andrew> except I've rebased onto current master.  I just wanted to make sure
Andrew> it was clear what the latest version of this patch was.

Thank you both for working on this.

Andrew> When used within gdb the above command produced output like this:
Andrew>   (gdb)
Andrew>   -echo-args a b c
Andrew>   ^done,args=["a","b","c"]
Andrew>   (gdb)

Not necessary for this patch but I wonder if there should also be a way
to emit an async event from Python.

Andrew> I have required that the fields names used in mi result output must
Andrew> follow C identifier restrictions (i.e. must match the regexp
Andrew> "[a-zA-Z][a-zA-Z0-9_]*").

I think '-' is also supported, e.g.:

      uiout->field_string ("thread-id", print_thread_id (thr));

Andrew> +@defun MICommand.invoke (arguments)
Andrew> +This method is called by @value{GDBN} when the new MI command is
Andrew> +invoked.
Andrew> +
Andrew> +@var{arguments} is a list of strings.  Note, that @code{--thread}
Andrew> +and @code{--frame} arguments are handled by @value{GDBN} itself therefore
Andrew> +they do not show up in @code{arguments}.

I guess this is done because an option may or may not take a value?
That kind of too bad since it would be nice if this could just be a
dictionary.

I suppose we could always also add an MI parsing helper, though Python
also has tons of things for this as well.

Andrew> +The strings used for @emph{VARIABLE} names in the @sc{GDB/MI} output
Andrew> +must follow the following rules; the string must be at least one
Andrew> +character long, the first character must be in the set
Andrew> +@code{[a-zA-Z]}, while every subsequent character must be in the set
Andrew> +@code{[a-zA-Z0-9_]}.

This should be updated to include '-'.

Andrew> +@smallexample
Andrew> +class MIEcho(gdb.MICommand):
Andrew> +    """Echo arguments passed to the command."""
Andrew> +
Andrew> +    def __init__(self, name, mode, toplevel = None):
Andrew> +        self._mode = mode
Andrew> +        super(MIEcho, self).__init__(name, toplevel)

What is 'toplevel' here?  It isn't documented for gdb.MICommand.

Andrew> +  /* Called when the MI command is invoked.  */
Andrew> +  virtual void do_invoke(struct mi_parse *parse) const override;

Andrew> +      if (key_repr != nullptr)
Andrew> +	key_repr_string = python_string_to_target_string (key_repr.get ());
Andrew> +      if (key_repr_string == nullptr)
Andrew> +	gdbpy_handle_exception ();
Andrew> +
Andrew> +      gdbpy_error (_("non-string object used as key: %s"),
Andrew> +		   key_repr_string.get ());

Someday I'd like gdb to solve the "denaturation" problem, where if
Python code throw some exception, it's turned into a generic exception
of some kind by gdb ... which if it is then caught again by Python, has
transmuted into something else.

Anyway this seems fine, just felt like I had to get that out.

Andrew> +    for (; *name != '\0'; ++name)
Andrew> +      if (!isalnum (*name) && *name != '_')
Andrew> +	return false;

'-'

Andrew> +  else if (PyIter_Check (result))
Andrew> +    {
Andrew> +      gdbpy_ref<> item;
Andrew> +      ui_out_emit_list list_emitter (uiout, field_name);
Andrew> +      while (true)
Andrew> +	{

Not for this patch but I wonder if emitting dictionaries via ui_out
might make for nicer CLI commands sometimes.

Andrew> +  gdbpy_enter enter_py (get_current_arch (), current_language);

Probably should just drop the parameters here.

thanks,
Tom