This change introduces diagnostic messages for panics. The intent is to be
able to use panics in situations where it is either not possible to or not
worth the time to recover from errors and ensure a consistent/sensible
system state. In those situations, we still ought to be able to provide the
user with useful information to attempt to get unstuck, since the error is
surely in response to some particular input, and maybe that input can be
tweaked to work around the problem.
Ideally, invalid states are avoided using the type system and statically
verified at compile-time. But this is not always possible, or in some cases
may be way more effort or cause way more code complexity than is worth,
given the unliklihood of the error occurring.
With that said, it's been interesting, over the past >10y that TAME has
existed, seeing how unlikely errors do sometimes pop up many years after
they were written. It's also interesting to have my intuition of what is
"unlikely" challenged, but hopefully it holds generally.
DEV-7145
I had previously used `Context` to hold the parser configuration for
repetition, since that was the easier option. But I now want to utilize the
`Context` for a stack for the superstate trampoline, and I don't want to
have to deal with the awkwardness of the repetition in doing so, since it
requires that the configuration be created during delegation, rather than
just being passed through to all child parsers.
This adds to a mess that needs cleaning up, but I'll do that after
everything is working.
DEV-7145
And here's the thing that I've been dreading, partly because of the
`macro_rules` issues involved. But, it's not too terrible.
This module was already large and complex, and this just adds to it---it's
in need of refactoring, but I want to be sure it's fully working and capable
of handling NIR before I go spending time refactoring only to undo it.
_This does not yet use trampolining in place of the call stack._ That'll
come next; I just wanted to get the macro updated, the superstate generated,
and tests passing. This does convert into the
superstate (`ParseState::Super`), but then converts back to the original
`ParseState` for BC with the existing composition-based delegation. That
will go away and will then use the equivalent of CPS, using the
superstate+`Parser` as a trampoline. This will require an explicit stack
via `Context`, like XIRF. And it will allow for tail calls, with respect to
parser delegation, if I decide it's worth doing.
The root problem is that source XML requires recursive parsing (for
expressions and statements like `<section>`), which results in recursive
data structures (`ParseState` enum variants). Resolving this with boxing is
not appropriate, because that puts heap indirection in an extremely hot code
path, and may also inhibit the aggressive optimizations that I need Rust to
perform to optimize away the majority of the lowering pipeline.
Once this is sorted out, this should be the last big thing for the
parser. This unfortunately has been a nagging and looming issue for months,
that I was hoping to avoid, and in retrospect that was naive.
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
Since we'll never be reading past the header, this is all that is needed.
If in the future this is violated, XIRF will cause a nice diagnostic error
displaying precisely what opening tag caused the increased level of nesting,
which will aid in debugging and allow us to determine if it ought to be
increased. Here's an example, if I set the max to `3`:
error: maximum XML element nesting depth of `3` exceeded
--> /home/.../foo.xmlo:261:10
|
261 | <preproc:sym-ref name=":_vproduct:vector_a"/>
| ^^^^^^^^^^^^^^^^ error: this opening tag increases the level of nesting past the limit of 3
Of course, the longer-term goal is to do away with `xmlo` entirely.
This had no (perceivable via `/usr/bin/time -v`, at least) impact on memory
or CPU time.
DEV-7145
"Mixed content" is the XML term representing element nodes mixed with text
nodes. For example, `foo <strong>bar</strong> baz` is mixed.
TAME supports text nodes as documentation, intended to be in a literate
style but never fully realized. In any case, we need to permit them, and I
wanted to do more than just ignore the nodes.
This takes a different approach than typical parser delegation---it has the
parent parser _preempt_ the child by intercepting text before delegation
takes place, rather than having the child reject the token (or possibly
interpret it itself!) and have to handle an error or dead state.
And while this makes it more confusing in terms of state machine stitching,
it does make sense, in the sense that the parent parser is really what
"owns" the text node---the parser is delegating _element_ parsing only, take
asserts authority when necessary to take back control where it shouldn't be
delegated.
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
Recent changes regarding whitespace were all to support this change (though
it was also needed for XIRF, pre- and post-root).
Now I'll have to conted with how I want to handle text nodes in various
circumstances, in terms of `ele_parse!`.
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
Whether or not quoting is appropriate depends on context, and that parent
context is already performing the quoting. For example:
error: expected `</rater>`, but found `<import>`
--> /home/[...]/foo.xml:2:1
|
2 | <rater xmlns="http://www.lovullo.com/rater"
| ------ note: element starts here
--> /home/[...]/foo.xml:7:3
|
7 | <import package="/rater/core/base" />
| ^^^^^^^ error: expected `</rater>`
In these cases (obviously I'm still working on the parser, since this is
nonsense), the parser is responsible for quoting the token "<import>".
DEV-7145
There were two problem errors: one showing "element element" and one showing
the value along with the name of the attribute.
The change for `<Attr as Display>::fmt` is debatable. I'm going to do this
for now (only show `@name`) and adjust later if necessary.
I'll need to go use `crate::fmt` consistently in previously-existing format
strings at some point, too.
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
The trace outputs a note in the footer indicating _why_ it's being output,
so that the reader understands both where the potentially-unexpected
behavior originates from and so they know (in the case of the feature flag)
how to inhibit it.
That information originally lived in `Parser`, where the `cfg` directive to
enable it lives, but it was moved into the abstraction. This corrects that.
DEV-7145
This has gotten large and was cluttering `feed_tok`. This also provides the
ability to more easily expand into other types of tracing in the future.
DEV-7145
This information is likely redundant in a lowering pipeline, but is more
useful outside of such a pipeline. It's also more clear.
`Object` does not implement `Display`, though, because that's too burdensome
for how it's currently used. Many `Object`s are also `Token`s though and,
if fed to another `Parser` for lowering, it'll get `Display::fmt`'d.
DEV-7145
Rust was warning that `cfg` was unused if both `test` and
`parser-trace-stderr`. This both allows that and adjusts the precedence to
make more sense for tests.
DEV-7145
Because of recovery, the trace otherwise paints a really confusing-looking
picture when given unexpected input.
This is large enough now that it really ought to be extracted from
`feed_tok`, but I'll wait to see how this evolves further. I considered
adding color too, but it's not yet clear to me that the visual noise will be
all that helpful.
DEV-7145
This flag allows toggling the parser trace that was previously only
available to tests. Unfortunately, at the time of writing, Cargo cannot
enable flags in profiles, so I have to check for either `test` or this flag
being set to enable relevant features.
This trace is useful as I start to run the parser against existing code
written in TAME so that our existing systems can help to guide my
development. Unlike the current tests, it also allows seeing real-world
data as part of the lowering pipeline, where multiple `Parser`s are in
play.
Having this feature flag also makes this feature more easily discoverable to
those wishing to observe how the lowering pipeline works.
DEV-7145
impl for `&Token` instead of Token; the writer is just copying data into the
destination stream anyway.
This will allow us to continue writing the token while also using it for
further processing, like `tee`.
DEV-7145
We need to be able to export generated identifiers. Trying to figure out a
syntax for this was a bit tricky considering how much is generated, so I
just settled on something that's reasonably clear and easy to parse with
`macro_rules!`.
I had intended to just make everything public by default and encapsulate
using private modules, but that then required making everything else that it
uses public (e.g. error and token objects), which would have been a bizarre
thing to do in e.g. test cases.
DEV-7145
Values can be parsed using `TryFrom<Attr>`. Previously only `From<Attr>`
was supported, which could not fail.
This is critical for parsing values into types, which will wrap `SymbolId`
to provide data assurances.
DEV-7145
The tests had certain things in scope, but now that I'm trying to use it
outside of those modules, some fixes are needed.
This is admittedly a sloppy commit, with a number of miscellaneous fixes. I
didn't bother separating it more because most of them are type fixes, and
the `From<Attr>` stuff is going to have to change into, likely,
`TryFrom<Attr>` so that parse failures can occur when attributes do not
match certain patterns.
DEV-7145
The only additional information needed was opening spans so that we can
provide useful information regarding closing tags.
This uses a generic Span in place of {Open,Close}Span because the latter
wasn't necessary, but more descriptive types would be nice; it may be
beneficial later on to introduce newtypes for each of the span generated by
{Open,Close}Span.
DEV-7145
This was a TODO for the attribute parser generator. The first attribute
will be kept and later ones will be ignored, producing an error. Recovery
permits further attribute parsing having ignored the duplicate.
DEV-7145
This allows an element to be repeated by the parent NT. The easiest way I
saw to implement this for now was to abuse the Context to provide a runtime
configuration that would allow the state machine to reset after it has
completed parsing.
This also influences error recovery, in that if we're expecting zero or more
of something, we cannot provide an error for an unexpected name, and instead
must emit a dead state so that the caller can determine what to do.
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
This resolves a TODO by including the name of the element whose attributes
are currently being parsed.
This also frees a parent from having to provide additional context, allowing
Display to be fully delegated when stitching.
DEV-7145
This introduces `Nt := (A | ... | Z);`, where `Nt` is the name of the
nonterminal and `A ... Z` are the inner nonterminals---it produces a parser
that provides a choice between a set of nonterminals.
This is implemented efficiently by understanding the QName that is accepted
by each of the inner nonterminals and delegating that token immediately to
the appropriate parser. This is a benefit of using a parser generator macro
over parser combinators---we do not need to implement backtracking by
letting inner parsers fail, because we know ahead of time exactly what
parser we need.
This _does not_ verify that each of the inner parsers accept a unique QName;
maybe at a later time I can figure out something for that. However, because
this compiles into a `match`, there is no ambiguity---like a PEG parser,
there is precedence in the face of an ambiguous token, and the first one
wins. Consequently, tests would surely fail, since the latter wouldn't be
able to be parsed.
This also demonstrates how we can have good error suggestions for this
parsing framework: because the inner nonterminals and their QNames are known
at compile time, error messages simply generate a list of QNames that are
expected.
The error recovery strategy is the same as previously noted, and subject to
the same concerns, though it may be more appropriate here: it is desirable
for the inner parser to fail rather than retrying, so that the sum parser is
able to fail and, once the Kleene operator is introduced, retry on another
potential element. But again, that recovery strategy may happen to work in
some cases, but'll fail miserably in others (e.g. placing an unknown element
at the head of a block that expects a sequence of elements would potentially
fail the entire block rather than just the invalid one). But more to come
on that later; it's not critical at this point. I need to get parsing
completed for TAME's input language.
DEV-7145
This adds the ability to bind identifiers to represent `OpenSpan` and
`CloseSpan`, available to the `@` and `/` maps. Since identifiers in TAME
originate from attributes, this may not get a whole lot of use, but it's
important to be available.
There is some awkwardness in that the opening span appears to be scoped to
the entire nonterminal, but it's actually only available in the `@`
mapping. I'll change this if it's actually needed; this keeps things simple
for now.
DEV-7145
Since the parsers produce streaming IRs, we need to be able to emit tokens
representing closing delimiters, where they are important.
This notably doesn't use spans; I'll add those next, since they're also
needed for the previous work.
DEV-7145
The comment explains the issue. I don't think the strategy is going to be a
desirable one, but I want to move on and observe in retrospect how it ought
to be handled.
The important part right now is that recovery is accounted for and possible,
which was a long-standing concern.
DEV-7145
This begins generating parsers that are capable of parsing elements. I need
to move on, so this abstraction isn't going to go as far as it could, but
let's see where it takes me.
This was the work that required the recent lookahead changes, which has been
detailed in previous commits.
This initial support is basic, but robust. It supports parsing elements
with attributes and children, but it does not yet support the equivalent of
the Kleene star (`*`). Such support will likely be added by supporting
parsers that are able to recurse on their own definition in tail position,
which will also require supporting parsers that do not add to the stack.
This generates parsers that, like all the other parsers, use enums to
provide a typed stack. Stitched parsers produce a nested stack that is
always bounded in size. Fortunately, expressions---which can nest
deeply---do not need to maintain ancestor context on the stack, and so this
should work fine; we can get away with this because XIRF ensures proper
nesting for us. Statements that _do_ need to maintain such context are not
nested.
This also does not yet support emitting an object on closing tag, which
will be necessary for NIR, which will be a streaming IR that is "near" to
the source XML in structure. This will then be used to lower into AIR for
the ASG, which gives structure needed for further analysis.
More information to come; I just want to get this committed to serve as a
mental synchronization point and clear my head, since I've been sitting on
these changes for so long and have to keep stashing them as I tumble down
rabbit holes covered in yak hair.
DEV-7145
Having the lookahead token generic over the `ParseState` was a pain in the
ass for stitching, since they shared the same token type but not the same
parser. I don't expect there to be any need to be able to infer other
parser-related types for a token of lookahead, so I'd rather just make my
life easier until such a thing is needed.
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
*NB: This is the initial change to introduce the token of lookahead, but this
does not fully integrate it. In particular, this is missing from the
stitching/delegation layer.*
This has been a long time coming, I suppose, though I had tried to avoid it
with `Parser::delegate_lookahead`. But the problem with doing that is that
it forced the ParserState to recurse, which both violates that I want no
looping constructs except for the toplevel, and performs additional stack
allocation as it is not in tail position.
The final straw was having to both return an error _and_ an aggregate object
for the attribute parser when an unexpected element is encountered (this
code is not yet committed). One option was to add a recovery object to the
error object, and formalize that, but then we have other concerns; for
example, what if that recovery object triggered an error? We'd have to mask
either the old or the new error. But we wouldn't want to mask either,
because the object causing the error would be the aggregate attributes,
which is _not_ a recovery object, but actual data we want to emit. And so
it's a kluge right off of the bat.
The use of a token of lookahaed is a more traditional approach and has uses
outside of just this one scenario. It'll also allow for the removal of
recursion from the existing ParserStates, and possibly the elimination of
dead state associated data, though I may end up leaving that; more to come.
Rust will also optimize away lookahead storage and processing in Parsers
that do not utilize it.
DEV-7145
I'd feel rather silly if I used `debug_assert!` for the sake of tests and
they weren't actually being run due to optimization settings.
This is just to catch potential future regressions; all is well today.
DEV-7145
There was only one test outside of the `parse` module using these
fields. The next commit will be introducing lookahead, and I do not want to
have to trust callers to ensure invariants are met.
DEV-7145
This reverts commit b973d36862.
Alright, I'm getting sick of fighting with myself on this. But rather than
just removing the last commit, I'm going to keep it around, so that my
thoughts are clearly documented for my future quarrels with myself.
Firstly: this added more overhead than I wanted it to. While it wasn't
significant, it did add 100--150ms to one of our largest systems, up from
~2.8s, which seems a bit much for a token that's really just meant to make
life easier for the parser.
Further, it seems that all I've managed to do is push my original problem to
a different layer---this started as a means to resolve having to emit both
an object and an error simultaneously in the case where aggregate attribute
parsing has completed, but we encounter an error on the next token (e.g. an
unexpected element). But XIRF, if it's missing AttrEnd, should throw an
error, but should also recover. Recovery is easy---just assume that it was
present---_but then we don't emit a XIRF `AttrEnd` token_, which is
necessary for downstream systems. So we'd need to either:
(a) emit both a token and an error; or
(b) panic.
But if we're doing (a), then the need for `AttrEnd` goes away, because it
solves the original problem (though the other concerns of the previous
commit still stand). (b) is not ideal at all, even though the missing token
does represent an internal system error; it's not something the user can
correct. But, given that it's something that the user cannot correct,
doesn't that imply that it's an awkward thing to include in the token
stream? So back to `AttrEnd` being an awkward PITA to have.
So, given (a), I'll just do that: errors will become more of a "hey, this
error just occurred, but I'm trying to recover---here's an object that you
should use if you choose to continue parsing, but it may or may not be what
you're looking for; proceed with caution". That flips the original script:
I imagined having external systems feed recovery tokens, but this
encapsulates recovery within the parser, which really is more appropriate,
though less flexible than having an omniscient external recovery system;
such a monolith was always an awkward concept and would be difficult to
implement cleanly.
This can also potentially be implemented as a generalization of the Dead
state change that allowed an object to be emitted alongside the
lookahead/error.
Anyway, back to where I was...I'm sure I'll look back on this in the future
shaking my head, reflecting on how naive I was.
DEV-7145
Mike Gerwitz
Corey Vollmer