This resolves the performance issues caused by Rust's failure to elide the
ElementStack (ArrayVec) memcpys on move.
Since XIRF is invoked tens of millions of times in some cases for larger
systems, prior to this change, failure to optimize away moves for XIRF
resulted in tens of millions of memcpys. This resulted in linking of one
program going from 1s -> ~15s. This change reduces it to ~2.5s with the
wip-xmlo-xir-reader flag on, with the extra time coming from elsewhere (the
subject of future changes).
In particular, this change introduces a new mutable reference to
`ParseState::parse_token`, which is a reference to a `Context` owned by the
caller (e.g. `Parser`). In the case of XIRF, this means that
`Parser<flat::State, _>` will own the `ElementStack`/`ArrayVec` instead of
`flat::State`; this allows the latter to remain pure and benefit from Rust's
move optimizations, without sacrificing the otherwise-pure implementation.
ParseStates that do not need a mutable context can use `NoContext` and
remain pure.
DEV-12024
This makes the necessary tweaks to have the entire linker work end-to-end
and produce a compatible xmle file (that is, identical except for
nondeterministic topological ordering). That's good, and finally that can
get off of my plate.
What's disappointing, and what I'll have more information on in future
commits, is how slow it is.
The linking of our largest package goes from ~1s -> ~15s with this
change. The reason is because of tens of millions of `memcpy` calls. Why?
The ParseState abstraction is pure and passes an owned `self` around, and
Parser replaces its own reference using this:
let result;
TransitionResult(Transition(self.state), result) =
take(&mut self.state).parse_token(tok);
Naively, this would store a copy of the old state in `result`, allocate a
new ParseState for `self.state`, pass the original or a copy to
`parse_token`, and then overwrite `self.state` with the new ParseState that
is returned once it is all over.
Of course, that'd be devastating. What we want to happen is for Rust to
realize that it can just pass a reference to `self.state` and perform no
copying at all.
For certain parsers, this is exactly what happens. Great!
But for XIRF, it we have this:
/// Stack of element [`QName`] and [`Span`] pairs,
/// representing the current level of nesting.
///
/// This storage is statically allocated,
/// allowing XIRF's parser to avoid memory allocation entirely.
type ElementStack<const MAX_DEPTH: usize> = ArrayVec<(QName, Span), MAX_DEPTH>;
/// XIRF document parser state.
///
/// This parser is a pushdown automaton that parses a single XML document.
#[derive(Debug, Default, PartialEq, Eq)]
pub enum State<const MAX_DEPTH: usize, SA = AttrParseState>
where
SA: FlatAttrParseState,
{
/// Document parsing has not yet begun.
#[default]
PreRoot,
/// Parsing nodes.
NodeExpected(ElementStack<MAX_DEPTH>),
/// Delegating to attribute parser.
AttrExpected(ElementStack<MAX_DEPTH>, SA),
/// End of document has been reached.
Done,
}
ParseState contains an ArrayVec, and its implementation details are causes
LLVM _not_ to elide the `memcpy`. And there's a lot of them.
Considering that ParseState is supposed to use only statically allocated
memory and be zero-copy, this is rather ironic.
Now, this _could_ be potentially fixed by not using ArrayVec; removing
it (and the corresponding checks for balanced tags) gets us down to
2s (which still needs improvement), but we can't have a core abstraction in
our system resting on a house of cards. What if the optimization changes
between releases and suddenly linking / building becomes shit slow? That's
too much of a risk.
Further, having to limit what abstractions we use just to appease the
compiler to optimize away moves is very restrictive.
The better option seems like to go back to what I used to do: pass around
`&mut self`. I had moved to an owned `self` to force consideration of _all_
state transitions, but I can try to do the same thing in a different type of
way using mutable references, and then we avoid this problem. The
abstraction isn't pure (in the functional sense) anymore, but it's safe and
isn't relying on delicate inlining and optimizer implementation details to
have a performant system.
More information to come.
DEV-10863
This concludes the bulk of the header parsing, though there are surely going
to be other issues when I try to read a real xmlo file, such as
whitespace. That is something I expect that I'd rather handle as part of
XIRF, but maybe I'll initially ignore it here just to get it working. We'll
see.
DEV-10863
This parses the symbol dependency list (adjacency list).
I'm noticing some glaring issues in error handling, particularly that the
token being parsed while an error occurs is not returned and so recovery is
impossible. I'll have to address that later on, after I get this parser
completed.
Another previous question that I had a hard time answering in prior months
was how I was going to compose boilerplate parsers, e.g. handling the
parsing of single-attribute elements and such. A pattern is clearly taking
shape, and with the composition of parsers more formalized, that'll be able
to be abstracted away. But again, that's going to wait until after this
parser is actually functioning. Too many delays so far.
DEV-10863
Ideally this would just be an attribute, but I guess I never got around to
making that change in the compiler and I don't want a detour right now.
DEV-10863
I clearly was not paying attention to what was correct behavior here, since
the tests also verified the wrong behavior: rather than taking the last
processed attribute span, we should be taking the span of the opening
tag for the `preproc:sym` node.
DEV-10863
This simply removes boilerplate.
This will receive concrete examples once I come up with docs for the entire
module; there's boilerplate involved in testing and documenting this in
isolation and the time investment is not worth it yet until I'm certain that
this will not be changed.
DEV-10863
This integrates much of the work done so far to parse into a
`XmloEvent::SymDecl`. The attribute parsing _is_ verbose, and I do intend
to abstract it away later on, but I'm going to wait on that for now.
The new reader should be finishing up soon, which is really exciting, since
I started working on this months ago (before having to take a break on
TAMER); I'm anticipating strong performance gains in the reader, and this is
a test that will tell us how the compiler will perform moving forward with
the abstractions that I've spent so much time on.
DEV-10863
This introduces a new method similar to the previous `delegate`, but with
another closure that allows for handling lookahead tokens from the child
parser.
Admittedly, this isn't exactly what I was going for---a list of arguments
isn't exactly self-documenting, especially with the brevity when the
arguments line up---but this was easy to do and so I'll run with this for
now.
This also modified `delegate` to accept a context, even though it wasn't
necessary, both for consistency with its lookup counterpart and for brevity
with the `into` argument (allowing, in our case, to just pass the name of
the variant, rather than a closure).
I'm not going to handle the actual starting and accepting state stitching
abstraction for now; I'd like to observe future boilerplate more before I
consider the best way to handle it, though I do have some ideas.
DEV-10863
This is the delegation portion of what I've come to call "state
stitching"---wiring together two state machines that recognize the same
input tokens.
This handles the delegation of tokens once the parser has been entered, but
does not yet handle the actual stitching part of it: wiring the start and
accepting states of the child parser to the parent.
This is indirectly tested by the XmloReader, but it will receive its own
tests once I further finalize this concept. I'm playing around with some
ideas. With that said, a quick visual inspection together with the
guarantees provided by the type system should convince any familiar reader
of its correctness.
DEV-10863
This wasn't the simplest thing to start with, but I wanted to explore
something with a higher level of complexity. There is some boilerplate to
observe here, including:
1. The state stitching (as I guess I'm calling it now) of SymtableState
with XmloReaderState is all boilerplate and requires no lookahead,
presenting an abstraction opportunity that I was holding off on
previously (attr parsing for XIRF requires lookahead).
2. This is simply collecting attributes into a struct. This can be
abstracted away in the future.
3. Creating stub parsers to verify that generics are stitched rather than
being tightly coupled with another state is boilerplate that maybe can
be abstracted away after a pattern is observed in future tests.
DEV-10863
This does some cleanup and adds `parse::Object` for use in disambiguating
`From` for `ParseStatus`, allowing the `Transition` API to be much more
flexible in the data it accepts and automatically converts. This allows us
to concisely provide raw output data to be wrapped, or provide `ParseStatus`
directly when more convenient.
There aren't yet examples in the docs; I'll do so once I make sure this API
is actually utilized as intended.
DEV-10863
This replaces u8 and will be used for the new XmloReader.
Previously I wasn't sure what direction TAMER was going to go in with
regards to dimensionality, but I do not expect that higher dimensions will
be supported, and if they are, they'd very likely compile down to lower ones
and create an illusion of higher-dimensionality.
Whatever the future holds, it's not used today, and I'd rather these types
be correct.
ASG needs changing too, but one step at a time.
DEV-10863
This converts the tuple type alias into a newtype, so that we may provide
our own implementations.
This differs from a previous approach that I took, which involved making
this type `Result<(S, T), (S, E)>` so that the return values composed well
with other functions. But the reality is that this is used only by other
`ParseState`s and `Parser`, so it's unnecessary.
However, this is also an attempt to utilize the new Try and FromResidual
traits; note how the Try associated types match precisely what I was trying
to do before, though they're used as intermediate types. I'll see how this
evolves.
DEV-10863
This allows the Results to compose and, importantly, is compatible with
`?` without having to put in any extra effort.
This makes puts the caller in an awkward spot, so I introduced a utility
function `result_tup0_invert` for now; we'll see if that stays or evolves
differently.
DEV-10863
Since this is the object produced by this parser, this is likely the most
useful first thing to present as a summary of what `XmloReader` actually
does.
DEV-10863
This removes the flag from most of the code, which also resolves the
indentation. Not only was it bothering me, but I don't want (a) every line
modified when the module body is hoisted and (b) `rustfmt` to reformat
everything when that happens.
This means that everything will be built, even though it's not used, when
the flag is off, but I see that as a good thing.
DEV-10863
Finally we get to do some actual parsing with all of the preparatory work!
This means that we're finally ready to fully replace the old XmloReader,
provided that I'm okay with some boilerplate / lack of abstractions for
now (and I am, because all I've been doing is working on abstractions to
prepare lowering operations).
DEV-10863
This makes more sense for pattern matching. Encapsulation of these fields
is not necessary, given that it's passed around as an owned value and its
`new` method constructs it verbatim; the individual fields are
self-validating.
DEV-10863
This introduces a WIP lowering operation, abstracting away quite a bit of
the manual wiring work, which is really important to providing an API that
provides the proper level of abstraction for actually understanding what the
system is doing.
This does not yet have tests associated with it---I had started, but it's a
lot of work and boilerplate for something that is going to
evolve. Generally, I wouldn't use that as an excuse, but the robust type
definitions in play, combined with the tiny amount of actual logic, provide
a pretty high level of confidence. It's very difficult to wire these types
together and produce something incorrect without doing something obviously
bad.
Similarly, I'm holding off on proper docs too, though I did write some
information here.
More to come, after I actually get to work on the XmloReader.
On a side note: I'm happy to have made progress on this, since this wiring
is something I've been dreading and wondering about since before the Parser
abstraction even existed.
Note also that this makes parser::feed_toks private again---I don't intend
to support push parsers yet, since they're only needed internally. Maybe
for error recovery, but I'll wait to decide until it's actually needed.
DEV-10863
This begins to transition XmloReader into a ParseState. Unlike previous
changes where ParseStates were composed into a single ParseState, this is
instead a lowering operation that will take the output of one Parser and
provide it to another.
The mess in ld::poc (...which still needs to be refactored and removed)
shows the concept, which will be abstracted away. This won't actually get
to the ASG in order to test that that this works with the
wip-xmlo-xir-reader flag on (development hasn't gotten that far yet), but
since it type-checks, it should conceptually work.
Wiring lowering operations together is something that I've been dreading for
months, but my approach of only abstracting after-the-fact has helped to
guide a sane approach for this. For some definition of "sane".
It's also worth noting that AsgBuilder will too become a ParseState
implemented as another lowering operation, so:
XIR -> XIRF -> XMLO -> ASG
These steps will all be streaming, with iteration happening only at the
topmost level. For this reason, it's important that ASG not be responsible
for doing that pull, and further we should propagate Parsed::Incomplete
rather than filtering it out and looping an indeterminate number of times
outside of the toplevel.
One final note: the choice of 64 for the maximum depth is entirely
arbitrary and should be more than generous; it'll be finalized at some point
in the future once I actually evaluate what maximum depth is reasonable
based on how the system is used, with some added growing room.
DEV-10863
This introduces a (still-private) way to _push_ tokens into the parser,
rather than relying purely on a pull-based interface. Not only does this
simplify the iterator, but this is also preparing to make the new `feed_tok`
public so that parsers can be composed in more contexts. I suspect that
this method may also be useful for error recovery, since it can be used to
inject tokens into arbitrary points of a token stream.
I kept the new method private for now so that I can introduce the new API
and docs separate from this refactoring.
DEV-10863
The parsing framework originally created for XIR is now more general and
useful to other things. We'll see how this evolves.
This needs additional documentation, but I'd like to see how it changes as
I implement XmloReader and then some of the source readers first.
DEV-10863
This adds a `Token` type to `ParseState`. Everything uses `xir::Token`
currently, but `XmloReader` will use `xir::flat::Object`.
Now that this has been generalized beyond XIR, the parser ought to be
hoisted up a level.
DEV-10863
This does a couple of things: it ensures that documents one and only one
root note, and it properly handles dead transitions once parsing is
complete (allowing it to be composed).
This should make XIRF feature-complete for the time being. It does rely on
the assumption that the reader is stripping out any trailing whitespace, so
I guess we'll see if that's true as we proceed.
DEV-10863
I'm not rendering errors yet in practice, so this wouldn't have been
noticed, but we want error messages to reference the final byte in a file on
EOF, not the offset of the last-encountered token, which would be confusing.
This doesn't _directly_ pertain to what I'm working on; I just happened to
notice it.
DEV-10863
XIRF introduced the concept of `Transition` to help document code and
provide mental synchronization points that make it easier to reason about
the system. I decided to hoist this into XIR's parser itself, and have
`parse_token` accept an owned state and require a new state to be returned,
utilizing `Transition`.
Together with the convenience methods introduced on `Transition` itself,
this produces much clearer code, as is evidenced by tree::Stack (XIRT's
parser). Passing an owned state is something that I had wanted to do
originally, but I thought it'd lead to more concise code to use a mutable
reference. Unfortunately, that concision lead to code that was much more
difficult than necessary to understand, and ended up having a net negative
benefit by leading to some more boilerplate for the nested types (granted,
that could have been alleviated in other ways).
This also opens up the possibility to do something that I wasn't able to
before, which was continue to abstract away parser composition by stitching
their state machines together. I don't know if this'll be done immediately,
but because the actual parsing operations are now able to compose
functionally without mutability getting the way, the previous state coupling
issues with the parent parser go away.
DEV-10863
This introduces XIR Flat (XIRF), which is conceptually between XIR and
XIRT. This provides a more appropriate level of abstraction for further
lowering operations to parse against, and removes the need for other parsers
to perform their own validations (inappropriately) to ensure well-formed
XML.
There is still some cleanup worth doing, including moving some of the
parsing responsibility up a level back into the XIR parser.
DEV-10863
This behavior is unchanged, but it allows us to create more constant spans
for testing. For example:
const S = DUMMY_SPAN.offset_add(1).unwrap();
This, in turn, will allow for removing lazy_static! for tests that use it
for span generation.
DEV-10863
The Options here are awkward and will be able to go away in the new reader
and in AsgBuilder once it has a proper state machine.
This gets rid of some of the initial migratory work for the new reader,
because PackageAttrs is gone. I'm going to wait to update this to the new
way until I get further into this.
DEV-11449
I'm finally back to TAMER development.
The original plan, some time ago, was to gate an entirely new XmloReader
behind a feature flag (wip-xmlo-xir-reader), and go from there, leaving the
existing implementation untouched. Unfortunately, it became too difficult
and confusing to marry the old aggregate API with the new streaming one.
AsgBuilder is the only system interacting with XmloReader, so I decided (see
previous commits) to just go the route of refactoring the existing
one. I'm not yet sure if I'll continue to progressively refactor this one
and eliminate the two separate implementations behind the flag, or if I'll
get this API similar and then keep the flag and reimplement it. But I'll
know soon.
DEV-11449
This is simply not worth it; the size is not going to be the bottleneck (at
least any time soon) and the generic not only pollutes all the things that
will use ASG in the near future, but is also incompatible with the SymbolId
default that is used everywhere; if we have to force it to 32 bits anyway,
then we may as well just default it right off the bat.
I thought that this seemed like a good idea at the time, and saving bits is
certainly tempting, but it was premature.
It's a bit odd that I've done next to nothing with TAMER for the past week
or so, and decided to do this one small thing before I go on break for the
holidays, but I felt compelled to do _something_. Besides, this gets me in
a better spot for the inevitable mental planning and writing I'll be doing
over the holidays.
This move was natural, given what this has evolved into---it has nothing to
do with the concept of a "tree", and the modules imports emphasized that
fact given the level of inappropriate nesting.
Now that the parser has been simplified by removing attributes, we can
further simplify the state transitions to make it more clear what further
refactoring can be done.
DEV-11339
More information can be found in the prior commit message, but I'll
summarize here.
This token was introduced to create a LL(0) parser---no tokens of
lookahead. This allowed the underlying TokenStream to be freely passed to
the next system that needed it.
Since then, Parser and ParseState were introduced, along with
ParseStatus::Dead, which introduces the concept of lookahead for a single
token---an LL(1) grammar.
I had always suspected that this would happen, given the awkwardness of
AttrEnd; it was just a matter of time before the right abstraction
manifested itself to handle lookahead.
DEV-11339
Note that AttrParse{r=>}State needs renaming, and Stack will get a better
name down the line too. This commit message is accurate, but confusing.
This performs the long-awaited task of trying to observe, concretely, how to
combine two automata. This has the effect of stitching together the state
machines, such that the union of the two is equivalent to the original
monolith.
The next step will be to abstract this away.
There are some important things to note here. First, this introduces a new
"dead" state concept, where here a dead state is defined as an _accepting_
state that has no state transitions for the given input token. This is more
strict than a dead state as defined in, for example, the Dragon Book, where
backtracking may occur.
The reason I chose for a Dead state to be accepting is simple: it represents
a lookahead situation. It says, "I don't know what this token is, but I've
done my job, so it may be useful in a parent context". The "I've done my
job" part is only applicable in an accepting state.
If the parser is _not_ in an accepting state, then an unknown token is
simply an error; we should _not_ try to backtrack or anything of the sort,
because we want only a single token of lookahead.
The reason this was done is because it's otherwise difficult to compose the
two parsers without requiring that AttrEnd exist in every XIR stream; this
has always been an awkward delimiter that was introduced to make the parser
LL(0), but I tried to compromise by saying that it was optional. Of course,
I knew that decision caused awkward inconsistencies, I had just hoped that
those inconsistencies wouldn't manifest in practical issues.
Well, now it did, and the benefits of AttrEnd that we had in the previous
construction do not exist in this one. Consequently, it makes more sense to
simply go from LL(0) to LL(1), which makes AttrEnd unnecessary, and a future
commit will remove it entirely.
All of this information will be documented, but I want to get further in
the implementation first to make sure I don't change course again and
therefore waste my time on docs.
DEV-11268
These were missed from a couple of commits ago, after I recalled that I
could now simplify the Stack variants; they were made more complicated due
to isolated attribute parsing.
These progressive refactorings do a good job illustrating why composing
parsers is better than a monolith---the complexity of the parsers is
significantly reduced, and the number of combinations of states are also
greatly reduced, which allows us to reason about them in isolation.
DEV-11268
This was added only for isolated attribute parsing. Of course, this does
mean that a new union type will be needed when combining the two parsers,
depending on the desired resolution, but that'll come at a later time and
possibly in a more general way.
DEV-11268
Mike Gerwitz