Re: [PATCH 10/28] qapi: Rework name checking in preparation of stricter checking

[Markus Armbruster]

John Snow <jsnow@redhat.com> writes:

> On 3/25/21 2:18 AM, Markus Armbruster wrote:
[...]
>> Apropos qapi-gen testing scripts.  I have scripts to show me how the
>> generated code changes along the way in a branch.  They evolved over a
>> long time, and try to cope with changes in the tree that are hardly
>> relevant anymore.  By now, they could quite possibly make Frankenstein
>> recoil in horror.
>> 
>
> Are they in the tree?

No, because in their current state, they are incomprehensible *and* need
frequent tinkering.

>                       Largely if the generated code changes it's 
> invisible to me, but I rely heavily on the unit tests. I guess maybe if 
> they are not in a state to upstream it might not be worth the hassle to 
> clean them, but I don't know.

The negative unit tests are fairly comprehensive, and guard against
screwing up error path reasonably well.

The positive unit tests compare the frontend state dumped by
test-qapi.py, and compile-test the generated code.  Reasonable
protection against frontend screwups.  Some protection against backend
screwups.  Plenty of unwanted code generation changes can go undetected.

A tool to show generated code changes for review is useful, and having
such a tool in the tree would be nice.

>> As a secondary purpose, the scripts show me how output of pycodestyle-3
>> and pylint change.  This would be uninteresting if the code in master
>> was clean against a useful configuration of these tools.  Your work has
>> been making it less interesting.
>> 
>
> --js