Tim's journal

I got my week off to a good start by restoring a "new" laptop from my TimeMachine backup, for the second time in about ten days. (After last week's 193 million memory errors.)

After the restore (which took about an hour, fortuitously coinciding with the weekly all-hands meeting) I started my three Rust workspaces building, and after some strategic sledgehammer (removing-the-build-directory) applications, that actually succeeded. So I went back to alternating between work on rustpkg and the typechecker refactor, as build times permitted.

I actually made some progress on rustpkg: last week I'd started trying to port an example of a custom Makefile into a package script instead. The example package I'm using, rust-sdl, currently has a Makefile with this rule (which only fires when building on Macs):

$(SDLXMAIN): $(SDL_PREFIX)/libSDLmain.a
        $(CP) $< $@
        $(AR) -x $@ SDLMain.o || $(RM) -f $@
        $(LD) -r SDLMain.o -o SDLXMain.o -alias _main _SDLX_main \
             -unexported_symbol main || $(RM) -f $@
        $(MV) SDLXMain.o SDLMain.o || $(RM) -f $@
        $(CHMOD) u+w $@ || $(RM) -f $@
        $(AR) -r $@ SDLMain.o || $(RM) -f $@

It's not for me to ask why; anyway, being fairly simple, this made a good example for me. I transliterated it into Rust like so:

const SDL_PREFIX : &static/str = "/usr/local/lib";

#[pkg_do(post_build)]
fn post_build() {
 // todo: delete new_sdl_lib_name if any commands fail
    let existing_sdl_lib_name = path(SDL_PREFIX).push("libSDLmain.a");
    let new_sdl_lib_name = path(SDL_PREFIX).push("libSDLXmain.a");
    let sdl_obj_file = ~"SDLMain.o";
    let sdlx_obj_file = ~"SDLXmain.o";
    copy_file(existing_sdl_lib_name, new_sdl_lib_name);
    archive_file(Extract, new_sdl_lib_name, sdl_obj_file);
    load_file(sdl_obj_file, sdlx_obj_file, ~[(~"_main", ~"_SDLX_main")],
              ~[(~"_main")]);
    move_file(sdlx_obj_file, sdl_obj_file);
    change_file_permissions((User, Writable), new_sdl_lib_name);
    archive_file(Extract, sdl_obj_file);
}

This is still sketchy and I haven't actually implemented most of the functions above yet. But the basic idea for what we're trying to do in rustpkg is that as much functionality as possible should be in the Rust build system; you shouldn't need to write a Makefile even to do something fancy.

So I spent that part of today trying to make this work. It doesn't yet, but I did get up to the point where rustpkg fails with a resolve error -- when calling out to compile the code in the package script -- because the functions I didn't implement yet aren't defined. Most of that time was spent defining a new version of rustpkg's PackageScript structure that has a PkgId field instead of a name field and a version field. PkgIds just bundle together the name and version of the package, both of which now will be inferred from the directory name (or version control info if present) if not declared. (The existing version of rustpkg that's up on github requires you to declare your package's name and version explicitly.)

The typechecker situation is sadder; I fixed my bad merges and ran tests, only to find that one of the run-fail tests was failing with a nondeterministic error:

• Assertion failed: ((Ty->isIntegerTy() || Ty->isPointerTy()) && "Invalid cast"), \
      function CreatePointerCast, \
      file /Users/tchevalier/rust3/src/llvm/lib/VMCore/Instructions.cpp, \
       line 2385.
  

• error: internal compiler error: Asked to compute contents of fictitious type

• rustc(29409,0x107cca000) malloc: *** error for object 0x7fdd9ba19f80: pointer \
      being freed was not allocated
  *** set a breakpoint in malloc_error_break to debug
  

• Assertion failed: (box->td != NULL), function free, \
     file /Users/tchevalier/rust3/src/rt/boxed_region.cpp, line 75.

These are all different error messages that I got on different runs of the same test. Because this behavior looked so similar to what was happening on the other, broken laptop (though with a different test), I thought maybe this one was borked too. What are the chances of that? It's like in the joke that goes "I always bring a bomb with me when I fly, because what are the chances of there being two bombs on the airplane?" Running valgrind, though, assuaged that fear, since it actually worked (and reported a metric crap-ton of invalid reads), as opposed to declaring its inability to execute rustc (like on the broken laptop). The errors suggest some sort of memory corruption having to do with types, which isn't totally surprising since rustc represents types in a totally unsafe way, and it would be easy to get something wrong. It's still puzzling me that I can get so many tests to work with only one failing if I introduced a bug like that, though. Usually, these kinds of failures are catastrophic.

Anyway, progress, in that I closed the day with a laptop that isn't obviously broken, unlike last Friday...

I succeeded at moving to Vancouver! A bit of culture shock. (Feel free to come up with a joke about moving out of a mutable reference or something.) I'm only here for less than three months, so I'm trying to absorb as much of Graydon's knowledge as possible in the meantime.

Continuing to work on the typechecker refactoring (yes, still); running up against very frustrating sporadic segfaults that I'm pretty sure are not to do with my changes, and that's about the most frustrating thing that can happen.

I also started working on rustpkg to finish up the work that our valiant volunteer Zack started. So far, trying to get rustpkg to build rust-sdl. Since rust-sdl requires some custom build logic -- currently implemented in a Makefile -- to rename the library file on Macs, I figured that might be a good starting point for porting that into build logic implemented in Rust that rustpkg can run. Or not; I still don't really know what I'm doing yet.

Figured I'd prioritize resuming regular posting ahead of making informative postings ;-)

I fixed the tests that weren't working yesterday in the typechecker refactoring branch, and I'm running the test suite again -- a few compile-fail tests have failed already, but those should be fun to figure out. (So far I see a few ICEs due to the fact that we're continuing typechecking in more places now, rather than returning early because of seeing an error type.)

My post from yesterday got a few more comments than usual! I'm always glad to get comments on my Rust posts, especially since it can be hard for people who don't use Dreamwidth regularly to comment on my blog, so I appreciate any effort towards that end. Here's how to comment if you don't have a Dreamwidth account.

Also along those lines, I'm well aware that my Rust posts swing wildly back and forth between explaining basic terms and assuming you know about things that you'd only know about if you'd read some obscure module in rustc::middle -- so if you're curious about anything, please don't assume you "should" know what it means and please do ask questions!

A few weeks ago, I agreed to be interviewed by a reporter for CNN Money (online only, there's no TV version of this as far as I know) about my financial situation -- the Transgender Law Center put me in touch with her. Of course, I would rather make the news for something other than being in massive debt, but I'm hoping this will help raise awareness of something like that. Here's the article, which goes with a slide show of me and five trans women, with each of our stories in our own words. I can't speak for any of the other five participants, but I was pleased with the level of accuracy with which Blake Ellis, the author, transcribed my words as you see them on the page.

Despite my net worth being in the negative five figures, I still try to donate to organizations I support, and (while doing my taxes for 2012) I had occasion to make a list of them. These are not the only organizations I've supported, but they are my favorites. Here they are, in case you have more money than you know what to do with and want suggestions.

Scarleteen

I wish I'd had access to this site when I was 12 or 13. (I still thought that "oral sex" referred to kissing at that point.) In a country where whether teenagers should get accurate information about sex in school is a controversial subject, sites like this are sorely needed. Scarleteen is a labor of love by Heather Corinna, whose online presence I've been following for a while now, and her commitment and dedication to maintaining the site for minimal reward is inspiring.

The Ada Initiative

Open-source and free software communities, as well as free culture projects like Wikipedia, continue to be hostile environments for women, people in gender, sexual and romantic minorities, and lots of other people who are from groups that have outsider status. It doesn't have to be that way, and it would be better for everyone if everyone who had desire and energy to contribute was able to participate in building the future without fear of humiliation. The Ada Initiative is the only group I know of that is specifically working to make that vision a reality. I have had the pleasure of meeting and interacting online with both of the founders, Mary Gardiner and Valerie Aurora, so I have particular confidence in TAI's effectiveness.

National Network of Abortion Funds

Incredibly, getting an abortion in the US in 2013 is still highly dependent on one's income, and we still are a nation where being forced to give birth to a baby because of inability to pay around $500 (or in many locations, thousands of dollars when travel costs are factored in) is far from unheard of. The National Network of Abortion Funds focuses specifically on funding abortions for people who can't afford them, as well as changing unjust laws like the Hyde Amendment. They are sometimes good (not perfect!) at using language in their publicity that acknowledges that people who get pregnant and need abortions aren't always women. Stacey Burns, the online communications manager for NNAF, friended me on Facebook after I did a Causes.com birthday wish for NNAF a few years back, and through reading her posts, I've gotten a good sense for the kind of activism that NNAF represents, and that it's something I want to support.

All Hands Volunteers

In the summer of 2010, I went to Léogâne, Haiti for six weeks to help with earthquake relief. All Hands Volunteers was the group I volunteered with. I ended up leaving Haiti after four weeks instead of six -- turns out heavy labor in extreme heat wasn't the thing I was best at (and after years and years of sitting at a desk 40+ hours a week -- who'd have guessed?) While I was there, though, I saw firsthand that All Hands is a group that's very effective at getting a lot of work done with a small group of very committed volunteers. Since then, they've initiated disaster relief projects in the Philippines as well as post-Sandy relief in Staten Island and Long Island. Some of the volunteers I met while working with All Hands were among the most inspiring people I've ever met.

Lyon-Martin Health Services

I'm biased -- Lyon-Martin, in San Francisco, is where I get my primary health care. It's a place where I can feel confident that I won't be treated awkwardly or be forced to educate about being a man with a transsexual body, and it's also a place where trans women, queer cis women, and genderqueer people can feel confident of the same. They operate on a shoestring and were close to shutting down not long ago. You should give not just if you want to support respectful health care in the Bay Area, but also if you want to make sure that the informed-consent model for trans health care spreads further.

Partners in Health

Years ago, I read _Mountains Beyond Mountains_ by Tracy Kidder (on the recommendation of a LiveJournal friend!), a biography of Paul Farmer -- who, along with Ophelia Dahl, founded Partners in Health -- and it has affected my thoughts, if not yet my actions, ever since. Paul Farmer's belief in and work towards providing the same health care to poor people that (say) a Silicon Valley software engineer like me would expect for themself or for their friends and family is challenging and is a source of hope. I also appreciate that while PIH is non-sectarian, it's inspired by liberation theology; _Mountains Beyond Mountains_ quotes Farmer as saying that he knew there had to be something to religion, because rich people hated it and poor people derived strength from it. I like that. And PIH gets stuff done (plus, Ophelia Dahl graduated from my alma mater, Wellesley, reminding me that not everyone from my school becomes an investment banker).

Transgender, Gender-Variant and Intersex Justice Project

There are big-name organizations that call themselves "LGBT" when the "T" really stands for transmisogyny and the B is silent, but the TGI Justice Project is the real thing. Just as PIH focuses on providing health care where it's needed the most, TGIJP focuses on the needs of that subset of the "LGBTIQ" cohort who need justice the most: trans women of color who are or have been incarcerated or who are targeted by the criminal justice system.

Having written this list, now I'm looking forward to having my debts paid off so I can support all of these organizations more thoroughly! If you particularly want to support organizations whose work is of a global nature, Partners in Health and All Hands Volunteers are your best bets on this list. Most of my favorites are US-centric, though, since I believe in helping with the needs that I'm most familiar with (since who else is going to but the people who are affected?)

I fixed #4736, which was easy enough, but that led to an adventure in refactoring the Rust typechecker.

Right now, most functions in the typechecker associate zero or more node IDs with types as a side effect (in a hash table that gets passed around), and return a bool. Every node in the abstract syntax tree (AST) has a node ID that identifies it uniquely.

The bool return value denotes whether or not the expression diverges. You can think of what that means like this: there's a primitive called fail, and fail or anything that definitively invokes fail diverges. (It's a little more complicated now because fail is a macro, but never you mind.) That information gets used to warn about statements that are statically known to be unreachable.

At the same time, one of the types that exists in the internal representation of types is ty_error, which represents an erroneous value. Normally a typechecker would abort at the first error it sees, but we want to report as many errors as possible during one compilation session, so we need a name for the type that erroneous subexpressions have. We need it because otherwise we end up reporting derived errors -- errors that are because ty_error doesn't match some other expected type. But if we saw a ty_error, we know we already reported an error about it, so we don't need to report lots more errors and overwhelm the programmer.

#4736 was because there was code that needed to execute to fill in certain data structures in error cases -- solely so we can keep typechecking and try to report more errors -- but it was getting skipped, reflecting an assumption that a certain error was fatal when it wasn't. At the same time, there were some missing checks for ty_error, so derived errors were getting reported.

But then I started thinking that I wanted to make the handling of ty_error more principled, so I changed the bool return value to a three-valued type that I called Outcome. An Outcome is either a ProperType (no errors), a BotType (saw something that diverges, which is not necessarily an error) or a TypeError (saw a type error). So in addition to the types that get written to tables as a side effect, typechecker functions now return an Outcome. This works, and is hopefully going to make it less likely for more derived errors to get reported accidentally.

But then I thought: why do we need a separate return flag in the first place? No function should ever return a type with a sub-component that is ty_error -- in that case, it should just return ty_error. For ty_bot, it's a little more complicated: for example, the expression true || fail!() doesn't always diverge (in fact, it never does) even though one of its components does. Nonetheless, I think if an expression isn't known to diverge, we don't need to know if any of its subcomponents are known to diverge.

The reason why the flag needs to be there right now is that the typing rules (as manifested in the typechecker) are non-compositional. We need the flag because we need to track information about an expression that isn't reflected in its type. But I think that's a bad code smell.

So I'm going to take a stab at making the rules compositional, which means that typechecker functions will return unit instead of bool or Outcome. This shouldn't be hard, because the internal representation of types already has a flag that says whether it contains ty_error -- so I can just make the constructors smart and return ty_error for anything whose representation would contain ty_error (like ~[ty_error]). It's a little harder to handle ty_bot correctly, but it shouldn't be too much harder.

I've submitted a proposal to the Open Source Bridge conference to give a tutorial talk on Rust. I can still edit the proposal between now and March 2, so I'd welcome any comments or suggestions about it!

Today I more or less only worked on #4678, and I finally got stage1 to compile! Of course, it doesn't work yet. So far, I've changed the ty param list in a trait item to be a non-empty list, so that it's obvious which param is the self type parameter. That's the point of the issue, making the Self type in a trait just like any other type parameter (because that's what it is, and this gets rid of some needless code). But, I haven't updated resolve yet to resolve the Self type to the first ty param. At least my build's failing in the way I expect it to.

And, unbelievably, I'm not done with one-tuples either; I got the compiler to compile and the whole test suite to pass except... the one test case that I added for one-tuples, which fails to parse in the exact same way it did in the first place. sadtrombone.wav

Well, maybe this meme is dying down, but I happened to re-read Emily's Trans 101, Upgoer Five Style and while it's good, I also couldn't resist writing my own version. Strangely enough, I felt like the limited vocabulary here helped me be clear, whereas when I was writing about my job, I felt too constrained by it in places (possibly because of having to circumlocute for technical terms, which was less of an issue here.)

---

Most people think that when a baby is very little, they can tell whether the baby is a boy or a girl. Also, they think that every baby is either a boy or a girl, never both.

They think this because they think that what a baby's between-the-legs looks like tells you whether the baby is a boy or a girl. But that's not true. Both boys and girls can have one kind of between-the-legs. And both boys and girls can have the other kind of between-the-legs. It's how you feel that makes you a boy or a girl, and babies can't talk to let other people know how they feel.

Also, there are more ways to feel than just boy or girl, and you can also feel any or all of those ways no matter what you have between your legs.

Some people think that everyone has a mark inside their cells that says they're a boy or that they're a girl. This, too, is wrong. These marks are real, but it's people who decided that one mark makes you a boy and the other mark makes you a girl. People are wrong sometimes.

Most people who get called a boy when they're a baby are boys, and most people who get called a girl when they're a baby are girls. It's harder for girls who got called boys, and boys who got called girls, and people who aren't boys or girls. There are two different ways in which it's harder.

First, some people have a picture of their body inside their brain that's of a body that looks and feels different than how the rest of their body is. You can't change that picture even by thinking very hard or getting help from another person. If you are this way, you have to change your body instead to make it match the picture in your brain.

Second, whether a person needs to change their body and does, or they need to change it but they can't, or they don't need to change it, many people aren't very nice to boys who don't look like they think a boy should look, or to girls who don't look like they think a girl should look, or when they can't decide if another person is a girl or a boy.

I'm in both the first and the second group. People thought I was a girl when I was a baby, but I was a boy. I didn't know this could even be true until I was much older. As soon as I found out that just because people thought I was a girl didn't mean I was one, I knew I wasn't a girl. After a while, I realized that I was a boy, and not someone who wasn't a girl or a boy. I was able to change my body to make it more like the picture that's built into my brain, so I'm much happier having a body now. And most people who see me realize I'm a boy without me having to tell them, which also makes me happy, because it was hard to explain to people who thought I was a girl that I was actually a boy.

It's harder for people who got called a boy when they were a baby but aren't boys, because lots of people are very afraid of people who they think have said no to being a boy. They think that someone else not wanting to look like a boy means being a boy won't be as fun for them. To deal with their fears about themselves, those people hurt other people. This gets in the way of the people who got called boys and aren't, who are just trying to live their lives.

You can make it better by believing people when they say that they're a girl, or that they're a boy, or that they're something else and not a girl or a boy. You can also make it better by telling people they are wrong when they make fun of others who they think are being boys wrong or being girls wrong.

Check it!

Triaged some bugs. It's now the case that I've reviewed open bugs 1-2000 in the past two weeks. That sector of the issue tracker is in pretty good order! Lots of the bugs that are still open are enhancements, the others mostly look pretty challenging. So it's not like we have a lot of old bugs we've rejected for no reason.

Following a FIXME, I tried to do what I thought would be a simple refactoring in resolve, and messed up trait method type parameters thoroughly. The usual.

Removed some more structural records, now that the Feb. 4 snapshot is finally live! But, it looks like I need yet another snapshot to remove the last of the structural records from the pipes macros. What's more, I don't understand the pipes compiler well enough to fix that, though I've been staring at it for quite some time today.

And that was today.

Snapshot: still not quite done yet! Rarrr. Suddenly, an obscure problem appeared as the result of the combination of my snapshot and another commit, probably involving borrow checking. That commit got backed out, Niko's trying to figure out what the problem is, and the results are still building.

In the meantime I'm trying to make tuple-like struct constructors work cross-crate, which involves diving back into metadata reading and parsing. Which is my favorite part of the compiler. Oh, no, not "favorite", the other one...

Slow day. I also learned that if I want to get to the office on time for the weekly 9 AM Rust meeting, I should really leave home *more* than an hour before.