[llvm-dev] RFC: Second draft of an LLVM Community Code of Conduct

Renato Golin via llvm-dev llvm-dev at lists.llvm.org
Tue Oct 20 04:07:43 PDT 2015

Hi Chandler,

I have two very specific and very important details I want to discuss
with you. The list of harassment concerns and the reporting policy
being unilateral, with the latter being much more important than the

I'm pleading people to avoid in-line reply, since the ideas are
complete on their own, and picking any one paragraph may lead to
misinterpretation of my words. Please, just read through, and comment
at the end of each block.

I also ask people not to interpret my words in any prejudicial way.
This is a mathematical analysis of the text and the community, and
it's the only type of analysis I'm physically able to make. If any of
this looks offensive, it's because I'm not using the "politically
correct" phrase construction and terms, not because I think human
beings should be treated differently based on any characteristic
irrelevant to the discussion. If you have any specific concern, let me
know in private, to avoid deviating us from the discussion at hand.

   0. Sources

There are two main ones: the Ada initiative and Sarah Sharp's blog,
both focus on gender inclusion. Personally, I agree with *all* of
that, but I don't think we should side with any specific concern.

Most of the time, prejudice is very similar, and advocating against
one type often equals most other types. But not always. A few

 * It's easy to spot gender by names, so *directed* prejudice on a
mailing list is easier to women (and ethnicity) than sexual
orientation of mental disability. Indirect prejudice is a lot easier
to spot overall.
 * Disability is well established, so it's very crass to discriminate
face to face, but some people feel more comfortable in discriminating
against women (or gay) in person, especially if surrounded by other
like minded men.
 * Skin colour and immigration status is a *very* contentious issue in
the US, but almost irrelevant in other places of the world. This
harder to spot electronically but *very* easy personally, also easier
to get it wrong in the latter.
 * Religious discrimination is not as common in US as it is in Europe,
because of the clash in cultural background in more mixed societies.
It's possible that Americans (and I mean North, Centre and South) are
more likely to be rude religiously than Europeans or Asians, due to
severe lack of exposure to non-Christian religions.
 * Some cultures think it's extremely disrespectful to treat women
like men. Others treat gun ownership like a religion, more important
than people's safety. Isn't a gun more physically dangerous than

While we could go on forever, the key issue here is that all of those
problems have infinite ramifications, but not all of them are treated
equally by people that suffer a specific type of prejudice.

I encourage you to draw another parallel here with me:

Ada's post brings the "community conduct" in the beginning, but focus
on physical contact, ie. conferences. In that area, discrimination is
very clear, sometimes aggressive, and therefore potentially dangerous.
Strong measures need to be taken quickly to safeguard the mental and
physical state of the other participants. But most of that has a lot
less practical value when dealing with mailing list or IRC conduct.

Sarah's post is a lot more inline with the electronic interactions,
which most of us do most of the year, and it doesn't bring much about
all the differences nor speaks on drastic terms. It's based on
respect. Plain and simple.

So, based on your own sources, physical encounters and measures *may*
need to be a lot more drastic than electronic ones, and that's not
reflected on your proposal.

   1. Harassment listing

My point, simply put, is that listing some types of harassment, but
not others, is prejudicial in itself against the people that face
harassment types not specifically listed.

Most of the codes you linked have a well separated block named
"Diversity Statement", which lists even more stuff than what you used.
CouchDB has both the list and the statement, which is highly

As I listed above, there are many cultural, social and ethnic issues
in each type of prejudice, and trying to list them all would be
impossible. So, I think we should stick to the basics, remembering
that this is an *international* community.

I was recently reading about the word "race" in the English language.
I see it as very derogative, trying to imply that we're *that*
different. But I take it this scientific take on the meaning of the
word may have been diluted over the years. I don't mind the use of it,
as long as all native English speaker folks agree it's conflict-free.

My proposal is to include one term for each *type* of conflict. I see:
 * "race" and/or "ethnicity" and/or "gender" as all conflicts
involving visual individual characteristics,
 * "sexual orientation" and/or "religion" and/or "political views" as
important non-visual characteristics,
 * "social status" and "disability" as important background
discriminatory issues

Everything else can be regarded as sub-groups of these. The codes you
linked all have three very interesting ones: genotype, phenotype,
neurotype. It would be possible to mathematically include *all* others
within those three, but that would be too terse.

So, I understand this is a spectrum, and that we have to find the
right threshold. Only, that threshold is different for each one of us,
because we all suffer from different types of prejudice.

One idea is to set on a finite list, say 10, and let people choose
which 10 they cannot live without, and pick the most common. After
all, this IS "including, but not limited to".

Or, if no one else is feeling segregated by that list, just keep it
and ignore everything I said. :)

   2. Unilateral reporting and actions

This is the critical one, and unlike the harassment listing, one that
I'll have *a lot* of trouble if it goes in this way.

The key points are:

 * There is *no* mention that the committee will gather, nor value,
the input of the accused.
 * Due to protection issues, the accused will most likely only be
contacted *after* the decision is taken
 * Most decisions are unilateral and cannot be appealed

If this was a form of government, it would be called despotism.
Without knowing what the committee will be, or how it'll be chosen,
replaced and be made responsible for their actions (some codes you
link account for some of that), there is little we can work with at
this moment. None of the codes nor the blog posts you linked have that
kind of unilateral and strict consequences. I'm not sure where you got
that from.

Factoring out immediate criminal danger, my proposal is simple:

 * The accused must be involved, and must be given *every* chance to
defend him/herself, including asking other people's statements,
publicly or privately,
 * The accused's own deficiencies *must* be taken into account when
presenting the verdict, and he/she should be consulted on which
response would be best,
 * *Every* decision must allow for appeal if new evidence is
presented, (the asymptote here is zero very early)
 * Any member of the community can request investigation of the
committee's decisions, by an independent representative panel.

By representative, I mean chosen by the people, for the people, and
accountable for their actions to the people, with the same vigour as
we deal with breaches of conduct.

Basically put, the ends don't justify the means. This reporting guide
puts too much power into a committee that hasn't even been decided how
it will be formed. Given how the LLVM Foundation was formed, I
understand why people are unsure how this will play out. However, even
if we did know exactly how the committee would be chosen, these
guarantees are still mandatory. As I said earlier, projects and people
change. If the code allows for abuse, given enough time, abuse will be


More information about the llvm-dev mailing list