This requires the name as part of the package definition, which in turn
removes a state (and all the combinations resulting from it) from
AirAggregate, which results in significant complexity reduction for a very
complex part of the system.
Pushing this complexity outward results in a reduction of overall
complexity, and obviates the question of where NIR will receive a generated
name.
DEV-13162
NOTE: This fixes the aforementioned commit that caused the linker to
temporarily fail (670c5d3a5d at time of
writing). This does introduce an extra forward slash into
`l:dep/preproc:sym/@src`, but that does not appear to cause any
problems. That will eventually go away, so I'm not going to bother with it
any further.
As the `xmlo` file is lowered into AIR, the name will be prefixed with a
leading slash (if necessary, which it is atm) and will emit an
`Air::BindIdent`.
This means that packages will be properly indexed by their canonical name on
load, which will be important when we share this with tamec.
DEV-13162
This modifies the xmlo reader, xmlo->AIR lowering, and AIR->ASG to introduce
a package for identifiers. It does not yet, however, add edges from the
package to the identifier.
Once edges are added, the DFS will change in undesirable ways, which will
require a new implementation. This is desirable to decouple from Petgraph
anyway, and then will be able to restore the prior single-pass sort+cycle
check.
That will also encapsulate visiting behavior within the `asg::graph` module
and, in turn, allow encapsulating `Asg.graph` finally.
DEV-13162
This doesn't go far enough, but it elaborates a bit---the existing was far
too much of a catch-all. It's important to take advantage of exhaustiveness
checks to ensure each transition is properly accounted for.
This parser is going to get more work over time, including right now, so I'm
not going to go too deep into this yet, but it's be useful (as a reader) to
compare it to e.g. asg::air's parsers' explicit enumeration of states and
favoring of explicit errors over dead state transitions.
DEV-13162
This invokes clippy as part of `make check` now, which I had previously
avoided doing (I'll elaborate on that below).
This commit represents the changes needed to resolve all the warnings
presented by clippy. Many changes have been made where I find the lints to
be useful and agreeable, but there are a number of lints, rationalized in
`src/lib.rs`, where I found the lints to be disagreeable. I have provided
rationale, primarily for those wondering why I desire to deviate from the
default lints, though it does feel backward to rationalize why certain lints
ought to be applied (the reverse should be true).
With that said, this did catch some legitimage issues, and it was also
helpful in getting some older code up-to-date with new language additions
that perhaps I used in new code but hadn't gone back and updated old code
for. My goal was to get clippy working without errors so that, in the
future, when others get into TAMER and are still getting used to Rust,
clippy is able to help guide them in the right direction.
One of the reasons I went without clippy for so long (though I admittedly
forgot I wasn't using it for a period of time) was because there were a
number of suggestions that I found disagreeable, and I didn't take the time
to go through them and determine what I wanted to follow. Furthermore, it
was hard to make that judgment when I was new to the language and lacked
the necessary experience to do so.
One thing I would like to comment further on is the use of `format!` with
`expect`, which is also what the diagnostic system convenience methods
do (which clippy does not cover). Because of all the work I've done trying
to understand Rust and looking at disassemblies and seeing what it
optimizes, I falsely assumed that Rust would convert such things into
conditionals in my otherwise-pure code...but apparently that's not the case,
when `format!` is involved.
I noticed that, after making the suggested fix with `get_ident`, Rust
proceeded to then inline it into each call site and then apply further
optimizations. It was also previously invoking the thread lock (for the
interner) unconditionally and invoking the `Display` implementation. That
is not at all what I intended for, despite knowing the eager semantics of
function calls in Rust.
Anyway, possibly more to come on that, I'm just tired of typing and need to
move on. I'll be returning to investigate further diagnostic messages soon.
This ASG implementation is a refactored form of original code from the
proof-of-concept linker, which was well before the span and diagnostic
implementations, and well before I knew for certain how I was going to solve
that problem.
This was quite the pain in the ass, but introduces spans to the AIR tokens
and graph so that we always have useful diagnostic information. With that
said, there are some important things to note:
1. Linker spans will originate from the `xmlo` files until we persist
spans to those object files during `tamec`'s compilation. But it's
better than nothing.
2. Some additional refactoring is still needed for consistency, e.g. use
of `SPair`.
3. This is just a preliminary introduction. More refactoring will come as
tamec is continued.
DEV-13041
This newtype allows a caller to prove (using types) that a parser of a given
type (`ParseState`) has been finalized.
This will be used by the lowering pipeline to ensure that all parsers in the
pipeline end up getting finalized (as you can see from a TODO added in the
code, one of them is missing). The lack of such a type was an oversight
during the (rather stressed) development of the parsing system, and I
shouldn't need to resort to unit tests to verify that parsers have been
finalized.
DEV-13158
This was accepting an early EOF when the active child `ParseState` was in an
accepting state, because it was not ensuring that anything on the stack was
also accepting.
Ideally, there should be nothing on the stack, and hopefully in the future
that's what happens. But with how things are today, it's important that, if
anything is on the stack, it is accepting.
Since `is_accepting` on the superstate is only called during finalization,
and because the check terminates early, and because the stack practically
speaking will only have a couple things on it max (unless we're in tail
position in a deeply nested tree, without TCO [yet]), this shouldn't be an
expensive check.
Implementing this did require that we expose `Context` to `is_accepting`,
which I had hoped to avoid having to do, but here we are.
DEV-7145
I'm disappointed that I keep having to implement features that I had hoped
to avoid implementing.
This introduces a "superstate" feature, which is intended really just to be
a sum type that is able to delegate to stitched `ParseState`s. This then
allows a `ParseState` to transition directly to another `ParseState` and
have the parent `ParseState` handle the delegation---a trampoline.
This issue naturally arises out of the recursive nature of parsing a TAME
XML document, where certain statements can be nested (like `<section>`), and
where expressions can be nested. I had gotten away with composition-based
delegation for now because `xmlo` headers do not have such nesting.
The composition-based approach falls flat for recursive structures. The
typical naive solution is boxing, which I cannot do, because not only is
this on an extremely hot code path, but I require that Rust be able to
deeply introspect and optimize away the lowering pipeline as much as
possible.
Many months ago, I figured that such a solution would require a trampoline,
as it typically does in stack-based languages, but I was hoping to avoid
it. Well, no longer; let's just get on with it.
This intends to implement trampolining in a `ParseState` that serves as that
sum type, rather than introducing it as yet another feature to `Parser`; the
latter would provide a more convenient API, but it would continue to bloat
`Parser` itself. Right now, only the element parser generator will require
use of this, so if it's needed beyond that, then I'll debate whether it's
worth providing a better abstraction. For now, the intent will be to use
the `Context` to store a stack that it can pop off of to restore the
previous `ParseState` before delegation.
DEV-7145
Previously a `Depth` was provided only for `Open` and `Close`. This depth
information, for example, will be used by NIR to quickly determine whether a
given parser ought to assert ownership of a text/comment token rather than
delegating it.
This involved modifying a number of test cases, but it's worth repeating in
these commits that this is intentional---I've been bit in the past using
`..` in contexts where I really do want to know if variant fields change so
that I can consider whether and how that change may affect the code
utilizing that variant.
DEV-7145
Various DUMMY_SPAN-derived spans are used by many test cases, so this
finally extracts them---something I've been meaning to do for some time.
This also places DUMMY_SPAN behind a `cfg(test)` directive to ensure that it
is _only_ used in tests; UNKNOWN_SPAN should be used when a span is actually
unknown, which may also be the case during development.
DEV-7145
This teaches XIRF to optionally refine Text into RefinedText, which
determines whether the given SymbolId represents entirely whitespace.
This is something I've been putting off for some time, but now that I'm
parsing source language for NIR, it is necessary, in that we can only permit
whitespace Text nodes in certain contexts.
The idea is to capture the most common whitespace as preinterned
symbols. Note that this heuristic ought to be determined from scanning a
codebase, which I haven't done yet; this is just an initial list.
The fallback is to look up the string associated with the SymbolId and
perform a linear scan, aborting on the first non-whitespace character. This
combination of checks should be sufficiently performant for now considering
that this is only being run on source files, which really are not all that
large. (They become large when template-expanded.) I'll optimize further
if I notice it show up during profiling.
This also frees XIR itself from being concerned by Whitespace. Initially I
had used quick-xml's whitespace trimming, but it messed up my span
calculations, and those were a pain in the ass to implement to begin with,
since I had to resort to pointer arithmetic. I'd rather avoid tweaking it.
tameld will not check for whitespace, since it's not important---xmlo files,
if malformed, are the fault of the compiler; we can ignore text nodes except
in the context of code fragments, where they are never whitespace (unless
that's also a compiler bug).
Onward and yonward.
DEV-7145
This produces useful parse traces that are output as part of a failing test
case. The parser generator macros can be a bit confusing to deal with when
things go wrong, so this helps to clarify matters.
This is _not_ intended to be machine-readable, but it does show that it
would be possible to generate machine-readable output to visualize the
entire lowering pipeline. Perhaps something for the future.
I left these inline in Parser::feed_tok because they help to elucidate what
is going on, just by reading what the trace would output---that is, it helps
to make the method more self-documenting, albeit a tad bit more
verbose. But with that said, it should probably be extracted at some point;
I don't want this to set a precedent where composition is feasible.
Here's an example from test cases:
[Parser::feed_tok] (input IR: XIRF)
| ==> Parser before tok is parsing attributes for `package`.
| | Attrs_(SutAttrsState_ { ___ctx: (QName(None, LocalPart(NCName(SymbolId(46 "package")))), OpenSpan(Span { len: 0, offset: 0, ctx: Context(SymbolId(1 "#!DUMMY")) }, 10)), ___done: false })
|
| ==> XIRF tok: `<unexpected>`
| | Open(QName(None, LocalPart(NCName(SymbolId(82 "unexpected")))), OpenSpan(Span { len: 0, offset: 1, ctx: Context(SymbolId(1 "#!DUMMY")) }, 10), Depth(1))
|
| ==> Parser after tok is expecting opening tag `<classify>`.
| | ChildA(Expecting_)
| | Lookahead: Some(Lookahead(Open(QName(None, LocalPart(NCName(SymbolId(82 "unexpected")))), OpenSpan(Span { len: 0, offset: 1, ctx: Context(SymbolId(1 "#!DUMMY")) }, 10), Depth(1))))
= note: this trace was output as a debugging aid because `cfg(test)`.
[Parser::feed_tok] (input IR: XIRF)
| ==> Parser before tok is expecting opening tag `<classify>`.
| | ChildA(Expecting_)
|
| ==> XIRF tok: `<unexpected>`
| | Open(QName(None, LocalPart(NCName(SymbolId(82 "unexpected")))), OpenSpan(Span { len: 0, offset: 1, ctx: Context(SymbolId(1 "#!DUMMY")) }, 10), Depth(1))
|
| ==> Parser after tok is attempting to recover by ignoring element with unexpected name `unexpected` (expected `classify`).
| | ChildA(RecoverEleIgnore_(QName(None, LocalPart(NCName(SymbolId(82 "unexpected")))), OpenSpan(Span { len: 0, offset: 1, ctx: Context(SymbolId(1 "#!DUMMY")) }, 10), Depth(1)))
| | Lookahead: None
= note: this trace was output as a debugging aid because `cfg(test)`.
DEV-7145
Oh what a tortured journey. I had originally tried to avoid formalizing
lookahead for all parsers by pretending that it was only needed for dead
state transitions (that is---states that have no transitions for a given
input token), but then I needed to yield information for aggregation. So I
added the ability to override the token for `Dead` to yield that, in
addition to the token. But then I also needed to yield lookahead for error
conditions. It was a mess that didn't make sense.
This eliminates `ParseStatus::Dead` entirely and fully integrates the
lookahead token in `Parser` that was previously implemented.
Notably, the lookahead token is encapsulated in `TransitionResult` and
unavailable to `ParseState` implementations, forcing them to rely on
`Parser` for recursion. This not only prevents `ParseState` from recursing,
but also simplifies delegation by removing the need to manually handle
tokens of lookahead.
The awkward case here is XIRT, which does not follow the streaming parsing
convention, because it was conceived before the parsing framework. It needs
to go away, but doing so right now would be a lot of work, so it has to
stick around for a little bit longer until the new parser generators can be
used instead. It is a persistent thorn in my side, going against the grain.
`Parser` will immediately recurse if it sees a token of lookahead with an
incomplete parse. This is because stitched parsers will frequently yield a
dead state indication when they're done parsing, and there's no use in
propagating an `Incomplete` status down the entire lowering pipeline. But,
that does mean that the toplevel is not the only thing recursing. _But_,
the behavior doesn't really change, in the sense that it would infinitely
recurse down the entire lowering stack (though there'd be an opportunity to
detect that). This should never happen with a correct parser, but it's not
worth the effort right now to try to force such a thing with Rust's type
system. Something like TLA+ is better suited here as an aid, but it
shouldn't be necessary with clear implementations and proper test
cases. Parser generators will also ensure such a thing cannot occur.
I had hoped to remove ParseStatus entirely in favor of Parsed, but there's a
lot of type inference that happens based on the fact that `ParseStatus` has
a `ParseState` type parameter; `Parsed` has only `Object`. It is desirable
for a public-facing `Parsed` to not be tied to `ParseState`, since consumers
need not be concerned with such a heavy type; however, we _do_ want that
heavy type internally, as it carries a lot of useful information that allows
for significant and powerful type inference, which in turn creates
expressive and convenient APIs.
DEV-7145
This isn't conceptally all that significant of a change, but there was a lot
of modify to get it working. I would generally separate this into a commit
for the implementation and another commit for the integration, but I decided
to keep things together.
This serves a role similar to AttrSpan---this allows deriving a span
representing the element name from a span representing the entire XIR
token. This will provide more useful context for errors---including the tag
delimiter(s) means that we care about the fact that an element is in that
position (as opposed to some other type of node) within the context of an
error. However, if we are expecting an element but take issue with the
element name itself, we want to place emphasis on that instead.
This also starts to consider the issue of span contexts---a blob of detached
data that is `Span` is useful for error context, but it's not useful for
manipulation or deriving additional information. For that, we need to
encode additional context, and this is an attempt at that.
I am interested in the concept of providing Spans that are guaranteed to
actually make sense---that are instantiated and manipulated with APIs that
ensure consistency. But such a thing buys us very little, practically
speaking, over what I have now for TAMER, and so I don't expect to actually
implement that for this project; I'll leave that for a personal
project. TAMER's already take a lot of my personal interests and it can
cause me a lot of grief sometimes (with regards to letting my aspirations
cause me more work).
DEV-7145
`ParseState` originally required `Default` for use with `mem::take` in
`Parser::feed_tok`. This unfortunately cannot last, since more specialized
parsers require context during initialization in order to provide useful
diagnostic information. (The other option is to require the caller to
augment errors with diagnostic information, but that would have to be
duplicated by every caller and complicates parser composition; I'd prefer
those diagnostic details remain encapsulated.)
Replacing `Default` with `Option` is uglier, but it ends up producing the
same assembly as `mem::take` did, at least at the time of writing. Because
Rust is able to elide unnecessary moves using this implementation, there is
no need for `unwrap_unchecked` or other unsafe methods, which is great,
since it shows that this parsing methodology is viable entirely in safe
Rust.
DEV-7145
Previously, `ParseStatus::Dead` always yielded
`ParseState::Token`. However, I'm working on introducing parsers that
aggregate (parsing XML attributes into structs), and those parsers do not
know that they have completed aggregation until they reach a dead state;
given that, I need to yield additional information at that time.
I played around with a number of alternative ideas, but this ended up being
the cleanest, relative to the effort involved. For example, introducing
another parameter to `ParseStatus::Dead` was too burdensome on APIs that
ought not concern themselves with the possibility of receiving an object in
addition to a lookahead token, since many parsers are not capable of doing
so (given that they map M:(N<=M)).
Another option that I abandoned fairly quickly was having
`is_accepting` (potentially renamed) return an aggregate object, since
that's on the side and didn't feel like it was part of the parsing pipeline.
The intent is to abstract this some in a new `ParseState` method for
delegation + aggregation.
DEV-7145
This replaces a tuple with a tuple struct that allows for calculating more
complete span information, such as the span encompassing the entire
attribute and the value span including the surrounding quotes.
This includes logic that ought to be abstracted into `Span` itself, and it's
not as formal as I'd like it to be (e.g. not ensuring context), but this is
a good starting point.
Note that parsers call `Token::span`, which in turn calculates the attribute
span, each time an attribute is encountered during lowering. But Rust does
a good job at optimizing away unnecessary operations, so this didn't have an
observable impact on time.
DEV-7145
This provides much more clarity as to what is going on. Further, it's less
ambiguous, since I'm about to introduce a new type of xmlo lowering into XIR
for writing the actual xmlo files.
DEV-7145
This finally uses `parse` all the way up to aggregation into the ASG, as can
be seen by the mess in `poc`. This will be further simplified---I just need
to get this committed so that I can mentally get it off my plate. I've been
separating this commit into smaller commits, but there's a point where it's
just not worth the effort anymore. I don't like making large changes such
as this one.
There is still work to do here. First, it's worth re-mentioning that
`poc` means "proof-of-concept", and represents things that still need a
proper home/abstraction.
Secondly, `poc` is retrieving the context of two parsers---`LowerContext`
and `Asg`. The latter is desirable, since it's the final aggregation point,
but the former needs to be eliminated; in particular, packages need to be
worked into the ASG so that `found` can be removed.
Recursively loading `xmlo` files still happens in `poc`, but the compiler
will need this as well. Once packages are on the ASG, along with their
state, that responsibility can be generalized as well.
That will then simplify lowering even further, to the point where hopefully
everything has the same shape (once final aggregation has an abstraction),
after which we can then create a final abstraction to concisely stitch
everything together. Right now, Rust isn't able to infer `S` for
`Lower<S, LS>`, which is unfortunate, but we'll be able to help it along
with a more explicit abstraction.
DEV-11864
This is intended to describe, to the user, the state that the parser is
in. This will be used to convey additional information for general parser
errors, but it should also probably be integrated into parsers' individual
errors as well when appropriate.
This is something I expected to add at some point, but I wanted to add them
because, when dealing with lowering errors, it can be difficult to tell
what parser the error originated from.
DEV-11864
This wraps `Ident` in a new `Object` variant and modifies `Asg` so that its
nodes are of type `Object`.
This unfortunately requires runtime type checking. Whether or not that's
worth alleviating in the future depends on a lot of different things, since
it'll require my own graph implementation, and I have to focus on other
things right now. Maybe it'll be worth it in the future.
Note that this also gets rid of some doc examples that simply aren't worth
maintaining as the API evolves.
DEV-11864
A previous commit mentioned that there's not a place for `Dim`, and
duplicated it between `asg` and `xmlo`. Well, `Dtype` is also needed in
both, and so here's a home for now.
`Dtype` has always been an inappropriate detail for the system and will one
day be removed entirely in favor of higher-level types; the machine
representation is up to the compiler to decide.
DEV-11864
asg_builder is about to be replaced, but in the process of simplifying the
destination IR (the ASG), I'm moving things into the proper place. This
never belonged here---it belongs with the actual lowering operation.
Previously, this was not reasoned about in terms of a lowering operation,
and was written when I was first introducing myself to Rust and trying to
get a proof-of-concept linker working.
DEV-11864
Previously, since the graph contained only identifiers, discovered roots
were stored in a separate vector and exposed to the caller. This not only
leaked details, but added complexity; this was left over from the
refactoring of the proof-of-concept linker some time ago.
This moves the root management into the ASG itself, mostly, with one item
being left over for now in the asg_builder (eligibility classifications).
There are two roots that were added automatically:
- __yield
- __worksheet
The former has been removed and is now expected to be explicitly mapped in
the return map, which is now enforced with an extern in `core/base`. This
is still special, in the sense that it is explicitly referenced by the
generated code, but there's nothing inherently special about it and I'll
continue to generalize it into oblivion in the future, such that the final
yield is just a convention.
`__worksheet` is the only symbol of type `IdentKind::Worksheet`, and so that
was generalized just as the meta and map entries were.
The goal in the future will be to have this more under the control of the
source language, and to consolodate individual roots under packages, so that
the _actual_ roots are few.
As far as the actual ASG goes: this introduces a single root node that is
used as the sole reference for reachability analysis and topological
sorting. The edges of that root node replace the vector that was removed.
DEV-11864
In the actual implementation (outside of tests), this is always looking up
before adding the symbol. This will simplify the API, while still retaining
errors, since the identifier will fail the state transition if the
identifier did not exist before attempting to set a fragment. So while this
is slower in microbenchmarks, this has no effect on real-world performance.
Further, I'm refactoring toward a streaming ASG aggregation, which is a lot
easier if we do not need to perform lookups in a separate step from the
ASG's primitives.
DEV-11864
This removes the generic on the Asg (which was formerly BaseAsg),
hard-coding `IdentObject`, which will further evolve. This makes the IR an
actual concrete IR rather than an abstract data structure.
These tests bring me back a bit, since they were written as I was still
becoming familiar with Rust.
DEV-11864
This is the beginning of an incremental refactoring to remove generics, to
simplify the ASG. When I initially wrote the linker, I wasn't sure what
direction I was going in, but I was also negatively influenced by more
traditional approaches to both design and unit testing.
If we're going to call the ASG an IR, then it needs to be one---if the core
of the IR is generic, then it's more like an abstract data structure than
anything. We can abstract around the IR to slice it up into components that
are a little easier to reason about and understand how responsibilities are
segregated.
DEV-11864
RSG (Ryan Specialty Group) recently announced a rename to Ryan Specialty (no
"Group"), but I'm not sure if the legal name has been changed yet or not, so
I'll wait on that.
These are no longer TODOs---they represent invalid tokens.
I'm going to put effort into providing further context with the diagnostic
system [right now] because these are internal errors caused by either
miscompilation or an incomplete reader.
DEV-10936
This is a working concept that will continue to evolve. I wanted to start
with some basic output before getting too carried away, since there's a lot
of potential here.
This is heavily influenced by Rust's helpful diagnostic messages, but will
take some time to realize a lot of the things that Rust does. The next step
will be to resolve line and column numbers, and then possibly include
snippets and underline spans, placing the labels alongside them. I need to
balance this work with everything else I have going on.
This is a large commit, but it converts the existing Error Display impls
into Diagnostic. This separation is a bit verbose, so I'll see how this
ends up evolving.
Diagnostics are tied to Error at the moment, but I imagine in the future
that any object would be able to describe itself, error or not, which would
be useful in the future both for the Summary Page and for query
functionality, to help developers understand the systems they are writing
using TAME.
Output is integrated into tameld only in this commit; I'll add tamec
next. Examples of what this outputs are available in the test cases in this
commit.
DEV-10935
This entirely removes the old XmloReader that has since been replaced with a
XIR-based reader.
I had been holding off on this because the new reader is slower, pending
performance optimizations (which I'll do a little later on), however the
performance loss is of no practical consideration and only affects the
linker, which is still fast.
Therefore, it's better to get this old code out of the way to simplify
refactoring going forward. In particular, I'm working on the diagnostic
system.
This is a little sad, in a way---this is some of my first Rust code that I'm
deleting.
DEV-10935
This does not deal directly with XIRF (that's composed into a pipeline
outside of this parser).
I'd like to clean up further...perhaps I should retire the
wip-xmlo-xir-reader flag now, despite the minor performance regression (see
previous recent commits for explanation).
DEV-10935
This aggregates all non-panic errors that can occur during link time, making
`Box<dyn Error>` unnecessary. I've been wanting to do this for a long time,
so it's nice seeing this come together. This is a powerful tool, in that we
know, at compile time, all errors that can occur, and properly report on
them and compose them. This method of error composition ensures that all
errors have a chance to be handled within their context, though it'll take
time to do so in a decent way.
This just maintains compatibility with the dynamic dispatch that was
previous occurring. This work is being done to introduce the initial
diagnostic system, which was really difficult/confusing to do without proper
errors types at the top level, considering the toplevel is responsible for
triggering the diagnostic reporting.
The cycle error is in particular going to be interesting once the system is
in place, especially once it provides spans in the future, since it will
guide the user through the code to understand how the cycle formed.
More to come.
DEV-10935
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
Mike Gerwitz