Re: [PATCH v2 06/12] qapi/source: Add builtin null-object sentinel

[Eduardo Habkost]

On Thu, Jan 14, 2021 at 02:39:35PM +0100, Markus Armbruster wrote:
> John Snow <jsnow@redhat.com> writes:
> 
> > On 1/13/21 10:39 AM, Markus Armbruster wrote:
> >> Spelling nitpick: s/builtin/built-in/ in the title.
> >> 
> >
> > Sure.
> >
> >> John Snow <jsnow@redhat.com> writes:
> >> 
> >>> We use None to represent an object that has no source information
> >>> because it's a builtin. This complicates interface typing, since many
> >>> interfaces expect that there is an info object available to print errors
> >>> with.
> >>>
> >>> Introduce a special QAPISourceInfo that represents these built-ins so
> >>> that if an error should so happen to occur relating to one of these
> >>> builtins that we will be able to print its information, and interface
> >>> typing becomes simpler: you will always have a source info object.
> >>>
> >>> This object will evaluate as False, so "if info" remains a valid
> >>> idiomatic construct.
> >>>
> >>> NB: It was intentional to not allow empty constructors or similar to
> >>> create "empty" source info objects; callers must explicitly invoke
> >>> 'builtin()' to pro-actively opt into using the sentinel. This should
> >>> prevent use-by-accident.
> >>>
> >>> Signed-off-by: John Snow <jsnow@redhat.com>
> >> 
> >> As I pointed out in review of v1, this patch has two aspects mixed up:
> >> 
> >> 1. Represent "no source info" as special QAPISourceInfo instead of
> >>     None
> >> 
> >> 2. On error with "no source info", don't crash.
> >> 
> >> The first one is what de-complicates interface typing.  It's clearly
> >> serving this patch series' stated purpose: "static typing conversion".
> >> 
> >> The second one is not.  It sidetracks us into a design discussion that
> >> isn't related to static typing.  Maybe it's something we should discuss.
> >> Maybe the discussion will make us conclude we want to do this.  But
> >> letting the static typing work get delayed by that discussion would be
> >> stupid, and I'll do what I can to prevent that.
> >> 
> >
> > It's not unrelated. It's about finding the most tactical incision to 
> > make the types as we actually use them correct from a static analysis 
> > context.
> >
> > Maybe there's another tactical incision to make that's "smaller", for 
> > some perception of "smaller", but it's not unrelated.
> 
> We don't have to debate, let alone agree on relatedness.
> 
> >> The stupidest possible solution that preserves the crash is adding an
> >> assertion right where it crashes before this patch: in
> >> QAPISourceInfo.__str__().  Yes, crashing in a __str__() method is not
> >> nice, but it's no worse than before.  Making it better than before is a
> >> good idea, and you're quite welcome to try, but please not in this
> >> series.  Add a TODO comment asking for "make it better", then sit on
> >> your hands.
> >
> > I'm recently back from a fairly long PTO, so forgive me if I am 
> > forgetting something, but I am not really sure I fundamentally 
> > understand the nature of this critique.
> >
> > Making functions not "crash" is a side-effect of making the types 
> > correct. I don't see it as scope-creep, it's a solution to a problem 
> > under active consideration.
> 
> I disagree.
> 
> The crash you "fix" is *intentional*.  I was too lazy to write something
> like
> 
>     assert self.info
> 
> and instead relied in self.info.whatever to crash.  I don't care how it
> crashes, as long as it does crash.
> 
> I *like* qapi-gen to crash on such internal errors.  It's easy, and
> makes "this is a bug, go report it" perfectly clear.
> 
> I'd also be fine with reporting "internal error, this is a bug, go
> report it".  Not in this series, unless it's utterly trivial, which I
> doubt.
> 
> I'm *not* fine with feeding made-up info objects to the user error
> reporting machinery without proof that it'll actually produce a useful
> error message.  Definitely not trivial, thus not in this series.

If you really don't want to change the existing behavior of the
code, I believe we have only two options:

1) Annotate self.info as QAPISourceInfo (not Optional),
   and add a hack to make the expression `self.info` crash if the
   argument to __init__() was None.

2) Annotate self.info as Optional[QAPISourceInfo], and adding
   manual asserts everywhere self.info is used.

Which of those two options do you find acceptable, Markus?

-- 
Eduardo