[libc-dev] [RFC] Ground truth for public API

Petr Hosek via libc-dev libc-dev at lists.llvm.org
Tue Oct 8 17:53:04 PDT 2019


Recently, there was a discussion (
http://lists.llvm.org/pipermail/llvm-dev/2019-July/133648.html) about build
dependency on Python and the conclusion of that thread is that it's
something we want to avoid.

Since the standard_api.py script is used for generating code (headers and
source), could we possibly use tablegen for that purpose (and describe the
API surface in a .td file)?

On Tue, Oct 8, 2019 at 2:23 PM Siva Chandra via libc-dev <
libc-dev at lists.llvm.org> wrote:

> Hello all,
>
> In the source tree, we have different items like header files,
> implementation .cpp file etc, which have to conform to the public API
> specified by various standards. Currently, the header files and
> implementation files are prepared in a disjointed fashion. That is, we
> write them manually and separately. There is no check to verify that
> these separate entities are coherent and that they are conforming to
> the standard. So, I propose fixing this using a ground truth file.
> Other files are either generated using the information found in this
> ground truth file, or are checked against this ground truth file. The
> different parts of this proposal are as follows:
>
> 1. Ground truth file syntax: There will be a single ground truth file
> for the entire public API surface of the libc. This file will be human
> and machine editable. I propose that it be a Python file named
> standard_api.py defining a dictionary in the following manner:
>
> API = {
>     <header file>: {
>         <func name>: {
>             "return_type": <return type>,
>             "args": [<arg list>],
>         }
>         ...
>     }
>     ...
> }
>
> 2. Generating all public headers: The current code has a few examples
> of generating header files. I propose that we generate all public
> header files using the information found in standard_api.py. This
> requires the addition of a new command to the header configuration
> language. Specifically, I propose adding a command named %%func. The
> usage of this command can be illustrated with the example of the round
> function:
>
>     // math.h.def
>     ...
>     %%func(round)
>     ...
>
> The header generator will replace the line on which the %%func command
> is listed with the declaration of the round function as specified in
> standard_api.py.
>
> 3. Entrypoint wrappers: The current way of adding a C symbol via a
> post processing step might not be feasible on all targets. For
> example: http://lists.llvm.org/pipermail/libc-dev/2019-October/000000.html
> .
> For such targets, I propose that we generate a C wrapper for the C++
> implementation as part of the add_entrypoint_obj rule. Availability of
> the ground truth file facilitates the generation of these wrappers in
> a straightforward manner. The wrappers also help in the use case of
> LTO across app + libc. Moreover, since the wrappers are generated from
> the ground truth, their successful compilation ensures that the
> implementation conforms to the ground truth.
>
> WDYT?
>
> Thanks,
> Siva Chandra
> _______________________________________________
> libc-dev mailing list
> libc-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/libc-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/libc-dev/attachments/20191008/ed141e8c/attachment.html>


More information about the libc-dev mailing list