Restore performance of parsing #206
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -340,7 +340,7 @@ function tryparsenext(tok, source, pos, len, b, code)::Tuple{Any, Int, UInt8, Re | |
| return val, pos, b, code | ||
| end | ||
|
|
||
| @inline function typeparser(::AbstractConf{T}, source::Union{AbstractVector{UInt8}, IO}, pos, len, b, code, pl, options) where {T <: Dates.TimeType} | ||
| fmt = options.dateformat | ||
| df = fmt === nothing ? default_format(T) : fmt | ||
| tokens = df.tokens | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -140,7 +140,7 @@ function typeparser(::AbstractConf{BigFloat}, source, pos, len, b, code, pl, opt | |
| end | ||
| end | ||
|
|
||
| @inline function typeparser(conf::AbstractConf{T}, source, pos, len, b, code, pl, options) where {T <: SupportedFloats} | ||
| # keep track of starting pos in case of invalid, we can rewind to start of parsing | ||
| startpos = pos | ||
| x = zero(T) | ||
|
|
@@ -298,7 +298,7 @@ getx(x, f) = f === nothing ? x : nothing | |
| @noinline _parsedigits(conf::AbstractConf{T}, source, pos, len, b, code, options, digits::IntType, neg::Bool, startpos, overflow_invalid::Bool, ndigits::Int, f::F) where {T, IntType, F} = | ||
| parsedigits(conf, source, pos, len, b, code, options, digits, neg, startpos, overflow_invalid, ndigits, f)::Tuple{rettype(T), ReturnCode, Int} | ||
|
|
||
| @inline function parsedigits(conf::AbstractConf{T}, source, pos, len, b, code, options, digits::IntType, neg::Bool, startpos, overflow_invalid::Bool=false, ndigits::Int=0, f::F=nothing) where {T, IntType, F} | ||
| x = zero(T) | ||
| anydigits = false | ||
| has_groupmark = _has_groupmark(options, code) | ||
|
|
@@ -312,7 +312,7 @@ function parsedigits(conf::AbstractConf{T}, source, pos, len, b, code, options, | |
| while true | ||
| if b <= 0x09 | ||
| if overflows(IntType) && digits > overflowval(IntType) | ||
| return _parsedigits(conf, source, pos, len, b + UInt8('0'), code, options, Base.inferencebarrier(_widen(digits)), neg, startpos, overflow_invalid, ndigits, f)::Tuple{rettype(T), ReturnCode, Int} | ||
| elseif ndigits > maxdigits(T) | ||
| # if input is way too big, just bail | ||
| fastseek!(source, startpos - 1) | ||
|
|
@@ -392,8 +392,11 @@ function parsedigits(conf::AbstractConf{T}, source, pos, len, b, code, options, | |
| # now we parse any digits following decimal point (if any); start `frac` at UInt64(0) | ||
| # `digits` still receives any fractional digits, `frac` just keeps track of how many digits | ||
| # were parsed to combine with any "e123" exponent numbers to determine final exponent value | ||
| if overflows(IntType) && digits > overflowval(IntType) | ||
| x, code, pos = _parsefrac(conf, source, pos, len, b, code, options, Base.inferencebarrier(_widen(digits)), neg, startpos, UInt64(0), overflow_invalid, ndigits, f)::Tuple{rettype(T), ReturnCode, Int} | ||
| else | ||
| x, code, pos = parsefrac(conf, source, pos, len, b, code, options, digits, neg, startpos, UInt64(0), overflow_invalid, ndigits, f) | ||
| end | ||
|
|
||
| @label done | ||
| return x, code, pos | ||
|
|
@@ -405,7 +408,7 @@ end | |
| @noinline _parsefrac(conf::AbstractConf{T}, source, pos, len, b, code, options, digits::IntType, neg::Bool, startpos, frac, overflow_invalid, ndigits, f::F) where {T, IntType, F} = | ||
| parsefrac(conf, source, pos, len, b, code, options, digits, neg, startpos, frac, overflow_invalid, ndigits, f)::Tuple{rettype(T), ReturnCode, Int} | ||
|
|
||
| @inline function parsefrac(conf::AbstractConf{T}, source, pos, len, b, code, options, digits::IntType, neg::Bool, startpos, frac, overflow_invalid, ndigits, f::F) where {T, IntType, F} | ||
| x = zero(T) | ||
| parsedanyfrac = false | ||
| FT = FLOAT64 | ||
|
|
@@ -434,7 +437,7 @@ function parsefrac(conf::AbstractConf{T}, source, pos, len, b, code, options, di | |
| b = peekbyte(source, pos) - UInt8('0') | ||
| b > 0x09 && break | ||
| if overflows(IntType) && digits > overflowval(IntType) | ||
| return _parsefrac(conf, source, pos, len, b + UInt8('0'), code, options, Base.inferencebarrier(_widen(digits)), neg, startpos, frac, overflow_invalid, ndigits, f)::Tuple{rettype(T), ReturnCode, Int} | ||
| end | ||
| end | ||
| b += UInt8('0') | ||
|
|
@@ -502,7 +505,7 @@ end | |
| @noinline _parseexp(conf::AbstractConf{T}, source, pos, len, b, code, options, digits, neg::Bool, startpos, frac, exp::ExpType, negexp, FT, overflow_invalid, ndigits, f::F) where {T, ExpType, F} = | ||
| parseexp(conf, source, pos, len, b, code, options, digits, neg, startpos, frac, exp, negexp, FT, overflow_invalid, ndigits, f)::Tuple{rettype(T), ReturnCode, Int} | ||
|
|
||
| @inline function parseexp(conf::AbstractConf{T}, source, pos, len, b, code, options, digits, neg::Bool, startpos, frac, exp::ExpType, negexp, FT, overflow_invalid, ndigits, f::F) where {T, ExpType, F} | ||
| x = zero(T) | ||
| # note that `b` has already had `b - UInt8('0')` applied to it for parseexp | ||
| while true | ||
|
|
@@ -535,7 +538,7 @@ function parseexp(conf::AbstractConf{T}, source, pos, len, b, code, options, dig | |
| @goto done | ||
| end | ||
| if overflows(ExpType) && exp > overflowval(ExpType) | ||
| return _parseexp(conf, source, pos, len, b, code, options, digits, neg, startpos, frac, Base.inferencebarrier(_widen(exp)), negexp, FT, overflow_invalid, ndigits, f)::Tuple{rettype(T), ReturnCode, Int} | ||
| end | ||
| end | ||
| @label done | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -4,7 +4,7 @@ overflowval(::Type{T}) where {T <: Integer} = div(typemax(T) - T(9), T(10)) | |
| # if we eventually support non-base 10 | ||
| # overflowval(::Type{T}, base) where {T <: Integer} = div(typemax(T) - base + 1, base) | ||
|
|
||
| @inline function typeparser(::AbstractConf{T}, source, pos, len, b, code, pl, opts) where {T <: Integer} | ||
| x = zero(T) | ||
| neg = false | ||
| has_groupmark = _has_groupmark(opts, code) | ||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change | ||||
|---|---|---|---|---|---|---|
|
|
@@ -2,7 +2,7 @@ isgreedy(::Type{T}) where {T <: AbstractString} = true | |||||
| isgreedy(::Type{Symbol}) = true | ||||||
| isgreedy(T) = false | ||||||
|
|
||||||
| @inline function typeparser(::AbstractConf{T}, source, pos, len, b, code, pl, opts) where {T <: AbstractString} | ||||||
|
Member
There was a problem hiding this comment. Inlining the String typeparser lets the String path flatten transitively into each pipeline specialization (typeparser → findendquoted → ...), which compounds the cache cost of the scanning-loop inlines;
Suggested change
[posted by claude] |
||||||
| if quoted(code) | ||||||
| code |= OK | ||||||
| return findendquoted(T, source, pos, len, b, code, pl, true, opts.cq, opts.e, opts.flags.stripquoted) | ||||||
|
|
||||||
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
findendquoted/finddelimiterare the two big scanning loops, and thequoted/delimiterlayers wrap every type's pipeline — marking them@inlinestamps a copy of the loop body into every (type × source × return-type) pipeline specialization the workload compiles. Keeping them as shared compiled units (together with the String typeparser suggestion) measured -1.4MB cache (-20%) and ~-0.3s precompile on this PR, with runtime parity: they do O(field-length) work per call, so the call into a shared type-stable instance amortizes — unlike the per-character float digit machine, where inlining is the right move.[posted by claude]