On Sun, Nov 19, 2023 at 2:23 PM Morten Brørup <m...@smartsharesystems.com> wrote: > > > From: Jerin Jacob [mailto:jerinjac...@gmail.com] > > Sent: Sunday, 19 November 2023 08.08 > > > > On Fri, Nov 17, 2023 at 1:57 PM Morten Brørup > > <m...@smartsharesystems.com> wrote: > > > > > > > From: Jerin Jacob [mailto:jerinjac...@gmail.com] > > > > Sent: Friday, 17 November 2023 05.34 > > > > > > > > On Thu, Sep 28, 2023 at 11:10 AM <jer...@marvell.com> wrote: > > > > > > > > > > From: Jerin Jacob <jer...@marvell.com> > > > > > > > > > > Define qualification criteria for external library > > > > > based on a techboard meeting minutes [1] and past > > > > > learnings from mailing list discussion. > > > > > > > > > > [1] > > > > > http://mails.dpdk.org/archives/dev/2019-June/135847.html > > > > > > > > > > Signed-off-by: Jerin Jacob <jer...@marvell.com> > > > > > > > > Ping for review and/or merge. > > > > > > > > > > > > > --- > > > > > v2: > > > > > - Added "Meson build integration" and "Code readability" > > sections. > > > > > > > > > > doc/guides/contributing/index.rst | 1 + > > > > > .../contributing/library_dependency.rst | 23 > > > > +++++++++++++++++++ > > > > > 2 files changed, 24 insertions(+) > > > > > create mode 100644 > > doc/guides/contributing/library_dependency.rst > > > > > > > > > > diff --git a/doc/guides/contributing/index.rst > > > > b/doc/guides/contributing/index.rst > > > > > index dcb9b1fbf0..e5a8c2b0a3 100644 > > > > > --- a/doc/guides/contributing/index.rst > > > > > +++ b/doc/guides/contributing/index.rst > > > > > @@ -15,6 +15,7 @@ Contributor's Guidelines > > > > > documentation > > > > > unit_test > > > > > new_library > > > > > + library_dependency > > > > > patches > > > > > vulnerability > > > > > stable > > > > > diff --git a/doc/guides/contributing/library_dependency.rst > > > > b/doc/guides/contributing/library_dependency.rst > > > > > new file mode 100644 > > > > > index 0000000000..687a3b6cef > > > > > --- /dev/null > > > > > +++ b/doc/guides/contributing/library_dependency.rst > > > > > @@ -0,0 +1,23 @@ > > > > > +.. SPDX-License-Identifier: BSD-3-Clause > > > > > + Copyright(c) 2023 Marvell. > > > > > + > > > > > +Library dependency > > > > > +================== > > > > > + > > > > > +This document defines the qualification criteria for external > > > > libraries that may be > > > > > +used as dependencies in DPDK drivers or libraries. > > > > > + > > > > > +- **Free availability**: The library must be freely available to > > > > build in either source or binary > > > > > + form, with a preference for source form. > > > > > > Suggest adding: > > > > > > - **Free use and distribution license**: The library must be freely > > available to use and distribute without any attached conditions. > > > > > > We must require a BSD-like license, to ensure that DPDK as a whole > > (including 3rd party libraries) remains BSD licensed, and can be used > > in commercial (closed source) applications. > > > > As far as I understand, The initial scope of was “free availability” > > for building. > > Free distribution is much wider scope. I don't think, current external > > libraries[1] have free distribution rights. > > [1] > > https://github.com/DPDK/dpdk/blob/main/doc/guides/gpus/cuda.rst > > https://gitlab.com/nvidia/headers/cuda-individual/cudart/- > > /blob/main/LICENSE?ref_type=heads > > I didn't mean the library source code; I only meant the library in binary > form. > > It is nice if the library's header files may be distributed too, but I don't > see it as a requirement, especially if they are freely available elsewhere > (preferably without imposing additional restrictions on the ability to use > and distribute the library in binary form). > > How about this instead: > > - **License permitting free use and distribution in binary form**: > The library's license must allow free and unconditional use and distribution > of the library in binary form.
Distribution and unconditional use is not the case for existing library dependencies such as https://gitlab.com/nvidia/headers/cuda-individual/cudart/-/blob/main/LICENSE?ref_type=heads So I am not sure, Which is the correct thing to do. Maybe we can discuss more in tech board meeting if there are no other comments in mailing list on this topic. > > We might want lawyers to verify the wording when we have agreed on our > intentions. > > > > > I am fine with either way, Feedback from others? > > > > > > > > > > + > > > > > +- **Compiler compatibility**: The library must be able to > > compile > > > > with a DPDK supported compiler > > > > > + for the given execution environment. For example, For Linux, > > the > > > > library must be able to compile > > > > > > Typo (after "For example,"): For -> for > > > > Ack. Will fix next version. > > > > > > > > > > + with GCC and/or clang. > > > > > + > > > > > +- **Documentation**: Must have adequate documentation for the > > steps > > > > to build it. > > > > > + > > > > > +- **Meson build integration**: The library must have standard > > method > > > > like ``pkg-config`` > > > > > + for seamless integration with DPDK's build environment. > > > > > + > > > > > +- **Code readability**: When the depended library is optional, > > use > > > > stubs to reduce the ``ifdef`` > > > > > + clutter to enable better code readability. > > > > > > Why does everyone keep insisting that stubs make code more readable? > > Sometimes #ifdef is better. > > > > Could you share a case where when #ifdefs is better(Just to understand > > the view).? > > If an external library provides some simple functions, and a group (i.e. a > subset) of functions in a DPDK library depends on that external library, then > it might be more readable if those functions are enabled/disabled as a group > in the DPDK library rather than individually. > > E.g. the external library provides functions for statistical processing, and > the DPDK library can be built with or without statistics, depending on using > the external library or not. If the DPDK library is built without statistics, > it should not register statistics availability towards a management module > (e.g. telemetry) if it is not really implemented within the DPDK library > (because the underlying functions are stubs). > > It might also be a matter of personal preferences. When reviewing some source > code, an #ifdef makes the availability of an underlying function perfectly > clear. Blindly calling a function (of an external library) doesn't really > show if the function is implemented for real, or just a stub. In general theme in DPDK code base that we are trying to avoid a lot of #ifdef in a given C file. Instead, we are doing following scheme. https://github.com/DPDK/dpdk/blob/main/drivers/ml/cnxk/meson.build#L84 https://github.com/DPDK/dpdk/blob/main/drivers/ml/cnxk/meson.build#L60 https://github.com/DPDK/dpdk/blob/main/drivers/ml/cnxk/mvtvm_ml_stubs.c > > My opinion on this might be tainted by my preference for building from > scratch. The distro people might see it very differently! Yeah. I don't have a strong opinion, I can change to following, if there are no comments on this. Or we can discuss more in TB meeting if there are no review comments in mailing list on this topic. **Code readability**: When the depended library is optional, use either stubs or ``#ifdef`` consistently, not a mix of both, to ensure code readability. > > > > > > > > > Please use something like this instead: > > > > > > - **Code readability**: When the depended library is optional, use > > either stubs or ``#ifdef`` consistently, not a mix of both, to ensure > > code readability. > > >
