Re: Automatic code formatting

On Sun, Jul 10, 2022 at 6:31 PM [off-list ref] wrote:

On July 10, 2022 8:59 PM, brian m. carlson wrote:

quoted

On 2022-07-10 at 22:13:01, rsbecker@nexbridge.com wrote:

quoted

Being one of the platforms that will be specifically excluded from
this proposal, I would like to offer an alternative. Before that,
please remember that not everything is Linux. My suggestion is to
create infrastructure to automatically format on add/commit. This
could be pluggable relatively simply with clean filter that is
language specific - perhaps with a helper option that installs the
formatter easily (because clean filters are notoriously painful to
install for newbies from my observations). It would be nice to have
something in perl that is more portable and pervasive than clang -
although perl could launch clang where available. I think having
infrastructure for code formatting that is built into git is actually
highly desirable - assuming that it is not unduly difficult to install
those. It would extend beyond git contributions, but the contributors
could be told (Contributor's Guide) that then need to follow standard
X, which may very well be clang format. There are java formatters, php
and perl formatters, even COBOL and TAL formatters. My position is
that having a standard way to plug these in is a more general plan
that would reach a larger community. Git contributions could then just
leverage standard git functionality.

I am willing to acknowledge the fact that not everybody has clang on their
preferred platform.  However, I assume you do have a laptop or desktop with
which you access your platform (since I believe it's a server
platform) and that you can install software there, or that you have the ability to
run some sort of virtualization framework on some system.

I am in general not very willing to say that we can't use or have useful developer
tools because of uncommon platforms.  Linux, Windows, macOS, and (I believe)
FreeBSD, NetBSD, and OpenBSD all support clang and related tools, and I don't
think it's unreasonable for us to expect that someone can have access to such a
system as part of development, even if that's in a VM.  Those six operating
systems plus Chrome OS constitute the overwhelming majority of desktop and
laptop systems, and there are several options which are free (both as in speech
and beer).

Moreover, clang and LLVM are extremely portable[0].  As a practical matter, any
platform wanting to support software written in Rust (a popular and growing
language) will likely need LLVM, and there is also a lot of software in C and C++
that only supports GCC-compatible compilers.  I do feel that providing support for
modern toolchains is an important part of providing a viable OS port, and that we
should be able to rely on porters for that OS to do that work.  I realize that LLVM is
not yet ported to your system, but I believe it's going to functionally need to
happen sooner or later.  When it does, you'll be able to send patches directly
without needing to copy to another OS to format the code.

I should point out that gcc will *never* according to our latest intel, be supported on my platforms. *Never* is, of course, an indeterminate definition, but until various matters I cannot legally discuss change, which are not likely for at least 5 years, anything depending on gcc is out of the picture and unavailable, including clang. I understand the position regarding git contributions, but I am trying to get the point across that formatting code to a standard is a more general desire than just git contributions. It is a broad desire/requirement that should be considered. Rather than making processes git-contribution-specific, providing a more general solution that git contributors can use is more effective.

I don't understand why this matters to the proposal, though.  My
experience with projects adopting code-formatting is that there are
always some people who just don't use it (e.g. because they use a
different IDE that doesn't integrate with it, or they never bother
installing the extra tool, etc.).  Such users run the risk of having
automated CI jobs flag their code with a problem, and then they have
to go back and fix it, but it's still a net win because reviewers
don't have to spend time catching those problems, and the original
folks would have had to fix their code anyway.  In fact, in some
projects, I've been one of those users and having the project use a
code formatter that I didn't want to bother to set up and run didn't
negatively impact me at all.

Whether or not we have an automatic formatter, you're still
responsible to fix your code to match the project's guidelines.  I
don't think that becomes any harder, because if we have an automatic
formatter, the _most_ that changes is there's a flag day where the
style is adjusted slightly.  After that point, all you have to do is
the same as today: look at the surrounding code to see the format and
follow it.  So, from what I can see, you lose nothing from this
proposal if it is implemented.