Compare commits

..

3 Commits

Author SHA1 Message Date
Cursor Agent d15f1d131f fix parser error highlight offsets for non-suffix slices
Co-authored-by: Thomas Pelletier <thomas@pelletier.dev>
2026-04-12 12:26:40 +00:00
Thomas Pelletier f36a3ece9e Reduce marshal and unmarshal overhead (#1044)
* Reduce marshal and unmarshal overhead

Targeted optimizations to reduce performance overhead introduced by
recent feature additions and the unsafe removal.

Unmarshal:
- parseKeyval: access the node directly in the builder's slice to set
  Raw, bypassing NodeAt which triggers a GC write barrier for the
  nodes-pointer on every key-value expression.
- Iterator.Next: cache the *nodes slice dereference in a local variable
  to avoid repeated pointer-to-slice indirection in the hot loop.

Marshal:
- Guard shouldOmitZero calls with an inlineable options.omitzero check.
  shouldOmitZero has inlining cost 1145 (budget 80), so avoiding the
  function call when omitzero is not set removes per-field overhead.
- Inline the isNil check in encodeMap. isNil has inlining cost 93
  (budget 80), so expanding it at the single hot call site avoids
  per-map-entry function call overhead.

Update README benchmarks.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-03-24 11:08:39 +00:00
Thomas Pelletier 77f3862df4 Fix benchmark script replacing internal package imports (#1042)
* Fix benchmark script replacing internal package imports

The sed command in bench() was replacing all occurrences of the go-toml
module path, including sub-package imports like internal/assert. This
caused the BurntSushi/toml benchmark to fail because it tried to import
github.com/BurntSushi/toml/internal/assert which doesn't exist.

Fix by anchoring the sed pattern to only match the import path when
followed by a closing quote, preserving internal package imports.

Also add a guard in the benchstathtml Python script to give a clear
error instead of an IndexError when no benchmark results are available.

https://claude.ai/code/session_016JGASo49PeFSfCaDxvrGFE

* Update benchmark results in README

https://claude.ai/code/session_016JGASo49PeFSfCaDxvrGFE

---------

Co-authored-by: Claude <noreply@anthropic.com>
2026-03-23 22:00:18 -04:00
11 changed files with 94 additions and 202 deletions
-25
View File
@@ -1,25 +0,0 @@
name: capabilities
on:
push:
branches:
- v2
pull_request:
branches:
- v2
jobs:
check:
name: check capabilities
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
with:
fetch-depth: 0
- name: Setup go
uses: actions/setup-go@v6
with:
go-version: "1.26"
- name: Install capslock
run: go install github.com/google/capslock/cmd/capslock@latest
- name: Check for new capabilities
run: ./caps.sh check
+1 -10
View File
@@ -53,14 +53,6 @@ go-toml is a TOML library for Go. The goal is to provide an easy-to-use and effi
- Commit messages must explain **why** the change is needed
- Keep messages clear and informative even if details are in the PR description
### Capabilities
go-toml tracks system-level capabilities using [capslock](https://github.com/google/capslock). The baseline is in `capability_baseline.txt` and CI enforces that it does not grow.
- **Do not introduce new capabilities.** PRs that increase the capability set (e.g., adding network access, subprocess execution, syscalls) are unlikely to be accepted.
- If a change causes the capabilities check to fail, do not update the baseline to make it pass. Instead, rethink the approach to avoid requiring new capabilities.
- To check locally: `./caps.sh check` (requires `capslock` installed via `go install github.com/google/capslock/cmd/capslock@latest`)
## Pull Request Checklist
Before submitting:
@@ -69,5 +61,4 @@ Before submitting:
2. No backward-incompatible changes (unless discussed)
3. Relevant documentation added/updated
4. No performance regression (verify with benchmarks)
5. Capabilities are not increasing (`./caps.sh check`)
6. Title is clear and understandable for changelog
5. Title is clear and understandable for changelog
-19
View File
@@ -180,25 +180,6 @@ description. Pull requests that lower performance will receive more scrutiny.
[benchstat]: https://pkg.go.dev/golang.org/x/perf/cmd/benchstat
### Capabilities
We use [capslock](https://github.com/google/capslock) to track what
system-level capabilities (file access, network, syscalls, etc.) each package
requires. The current baseline is in `capability_baseline.txt`. CI will fail if
a change introduces a new capability.
**Pull requests that increase the set of capabilities are unlikely to be
accepted.** go-toml is a parsing library and should not need network access,
subprocess execution, or other capabilities beyond what it already uses.
If you believe a new capability is genuinely needed, discuss it in an issue
first. To update the baseline after approval:
```bash
go install github.com/google/capslock/cmd/capslock@latest
./caps.sh generate
```
### Style
Try to look around and follow the same format and structure as the rest of the
+27 -27
View File
@@ -235,17 +235,17 @@ the AST level. See https://pkg.go.dev/github.com/pelletier/go-toml/v2/unstable.
Execution time speedup compared to other Go TOML libraries:
<table>
<thead>
<tr><th>Benchmark</th><th>go-toml v1</th><th>BurntSushi/toml</th></tr>
</thead>
<tbody>
<tr><td>Marshal/HugoFrontMatter-2</td><td>1.9x</td><td>2.2x</td></tr>
<tr><td>Marshal/ReferenceFile/map-2</td><td>1.7x</td><td>2.1x</td></tr>
<tr><td>Marshal/ReferenceFile/struct-2</td><td>2.2x</td><td>3.0x</td></tr>
<tr><td>Unmarshal/HugoFrontMatter-2</td><td>2.9x</td><td>2.7x</td></tr>
<tr><td>Unmarshal/ReferenceFile/map-2</td><td>2.6x</td><td>2.7x</td></tr>
<tr><td>Unmarshal/ReferenceFile/struct-2</td><td>4.6x</td><td>5.1x</td></tr>
</tbody>
<thead>
<tr><th>Benchmark</th><th>go-toml v1</th><th>BurntSushi/toml</th></tr>
</thead>
<tbody>
<tr><td>Marshal/HugoFrontMatter-2</td><td>2.1x</td><td>2.0x</td></tr>
<tr><td>Marshal/ReferenceFile/map-2</td><td>2.0x</td><td>2.0x</td></tr>
<tr><td>Marshal/ReferenceFile/struct-2</td><td>2.3x</td><td>2.5x</td></tr>
<tr><td>Unmarshal/HugoFrontMatter-2</td><td>3.3x</td><td>2.8x</td></tr>
<tr><td>Unmarshal/ReferenceFile/map-2</td><td>2.9x</td><td>3.0x</td></tr>
<tr><td>Unmarshal/ReferenceFile/struct-2</td><td>4.8x</td><td>5.0x</td></tr>
</tbody>
</table>
<details><summary>See more</summary>
<p>The table above has the results of the most common use-cases. The table below
@@ -253,22 +253,22 @@ contains the results of all benchmarks, including unrealistic ones. It is
provided for completeness.</p>
<table>
<thead>
<tr><th>Benchmark</th><th>go-toml v1</th><th>BurntSushi/toml</th></tr>
</thead>
<tbody>
<tr><td>Marshal/SimpleDocument/map-2</td><td>1.8x</td><td>2.7x</td></tr>
<tr><td>Marshal/SimpleDocument/struct-2</td><td>2.7x</td><td>3.8x</td></tr>
<tr><td>Unmarshal/SimpleDocument/map-2</td><td>3.8x</td><td>3.0x</td></tr>
<tr><td>Unmarshal/SimpleDocument/struct-2</td><td>5.6x</td><td>4.1x</td></tr>
<tr><td>UnmarshalDataset/example-2</td><td>3.0x</td><td>3.2x</td></tr>
<tr><td>UnmarshalDataset/code-2</td><td>2.3x</td><td>2.9x</td></tr>
<tr><td>UnmarshalDataset/twitter-2</td><td>2.6x</td><td>2.7x</td></tr>
<tr><td>UnmarshalDataset/citm_catalog-2</td><td>2.2x</td><td>2.3x</td></tr>
<tr><td>UnmarshalDataset/canada-2</td><td>1.8x</td><td>1.5x</td></tr>
<tr><td>UnmarshalDataset/config-2</td><td>4.1x</td><td>2.9x</td></tr>
<tr><td>geomean</td><td>2.7x</td><td>2.8x</td></tr>
</tbody>
<thead>
<tr><th>Benchmark</th><th>go-toml v1</th><th>BurntSushi/toml</th></tr>
</thead>
<tbody>
<tr><td>Marshal/SimpleDocument/map-2</td><td>2.0x</td><td>2.9x</td></tr>
<tr><td>Marshal/SimpleDocument/struct-2</td><td>2.5x</td><td>3.6x</td></tr>
<tr><td>Unmarshal/SimpleDocument/map-2</td><td>4.2x</td><td>3.4x</td></tr>
<tr><td>Unmarshal/SimpleDocument/struct-2</td><td>5.9x</td><td>4.4x</td></tr>
<tr><td>UnmarshalDataset/example-2</td><td>3.2x</td><td>2.9x</td></tr>
<tr><td>UnmarshalDataset/code-2</td><td>2.4x</td><td>2.8x</td></tr>
<tr><td>UnmarshalDataset/twitter-2</td><td>2.7x</td><td>2.5x</td></tr>
<tr><td>UnmarshalDataset/citm_catalog-2</td><td>2.3x</td><td>2.3x</td></tr>
<tr><td>UnmarshalDataset/canada-2</td><td>1.9x</td><td>1.5x</td></tr>
<tr><td>UnmarshalDataset/config-2</td><td>5.4x</td><td>3.0x</td></tr>
<tr><td>geomean</td><td>2.9x</td><td>2.8x</td></tr>
</tbody>
</table>
<p>This table can be generated with <code>./ci.sh benchmark -a -html</code>.</p>
</details>
-1
View File
@@ -1 +0,0 @@
github.com/pelletier/go-toml/v2: CAPABILITY_REFLECT, CAPABILITY_UNANALYZED, CAPABILITY_UNSAFE_POINTER
-101
View File
@@ -1,101 +0,0 @@
#!/usr/bin/env bash
#
# Generates or checks the capability baseline for go-toml.
#
# Usage:
# ./caps.sh generate # regenerate capability_baseline.txt
# ./caps.sh check # check that capabilities haven't grown
#
# Requires: go, capslock (go install github.com/google/capslock/cmd/capslock@latest)
set -euo pipefail
BASELINE="capability_baseline.txt"
CAPSLOCK="${CAPSLOCK:-capslock}"
# Capabilities that must never appear in any package.
FORBIDDEN_CAPS=(
CAPABILITY_NETWORK
CAPABILITY_CGO
CAPABILITY_EXEC
)
capslock_to_baseline() {
"$CAPSLOCK" -packages=. -output=package -granularity=package \
| jq -r 'to_entries | sort_by(.key) | .[] | .key + ": " + (.value | sort | join(", "))'
}
generate() {
capslock_to_baseline > "$BASELINE"
echo "Wrote $BASELINE"
}
check() {
if [ ! -f "$BASELINE" ]; then
echo "ERROR: $BASELINE not found. Run '$0 generate' first."
exit 1
fi
current=$(mktemp)
trap 'rm -f "$current"' EXIT
capslock_to_baseline > "$current"
failed=0
# Check for forbidden capabilities in current output.
for cap in "${FORBIDDEN_CAPS[@]}"; do
if grep -q "$cap" "$current"; then
echo "FORBIDDEN capability found: $cap"
grep "$cap" "$current"
failed=1
fi
done
# Extract all unique capability names from baseline and current.
baseline_caps=$(grep -oE 'CAPABILITY_[A-Z_]+' "$BASELINE" | sort -u)
current_caps=$(grep -oE 'CAPABILITY_[A-Z_]+' "$current" | sort -u)
# Check for new capability names not in the baseline.
new_caps=$(comm -13 <(echo "$baseline_caps") <(echo "$current_caps"))
if [ -n "$new_caps" ]; then
echo "NEW capabilities detected (not in baseline):"
echo "$new_caps"
failed=1
fi
# Check for new per-package capabilities (a package gained a capability it didn't have before).
while IFS=': ' read -r pkg caps; do
baseline_pkg_caps=$(grep "^${pkg}:" "$BASELINE" 2>/dev/null | sed 's/^[^:]*: //' || true)
if [ -z "$baseline_pkg_caps" ]; then
echo "NEW package with capabilities: $pkg: $caps"
failed=1
continue
fi
# Check each capability in current for this package
for cap in $(echo "$caps" | tr ', ' '\n' | grep -v '^$'); do
if ! echo "$baseline_pkg_caps" | grep -q "$cap"; then
echo "NEW capability for $pkg: $cap"
failed=1
fi
done
done < "$current"
if [ "$failed" -eq 1 ]; then
echo ""
echo "FAILED: capabilities have grown."
echo "If this is intentional, run '$0 generate' and commit the updated $BASELINE."
exit 1
fi
echo "OK: no new capabilities detected."
}
case "${1:-}" in
generate) generate ;;
check) check ;;
*)
echo "Usage: $0 {generate|check}"
exit 1
;;
esac
+6 -1
View File
@@ -147,7 +147,7 @@ bench() {
pushd "$dir"
if [ "${replace}" != "" ]; then
find ./benchmark/ -iname '*.go' -exec sed -i -E "s|github.com/pelletier/go-toml/v2|${replace}|g" {} \;
find ./benchmark/ -iname '*.go' -exec sed -i -E "s|github.com/pelletier/go-toml/v2\"|${replace}\"|g" {} \;
go get "${replace}"
fi
@@ -195,6 +195,11 @@ for line in reversed(lines[2:]):
"%.1fx" % (float(line[3])/v2), # v1
"%.1fx" % (float(line[7])/v2), # bs
])
if not results:
print("No benchmark results to display.", file=sys.stderr)
sys.exit(1)
# move geomean to the end
results.append(results[0])
del results[0]
+21
View File
@@ -286,6 +286,27 @@ func TestDecodeError_Position(t *testing.T) {
}
}
func TestDecodeError_InvalidKeyStartAfterComment(t *testing.T) {
doc := "# comment\n= \"value\""
var out map[string]string
err := Unmarshal([]byte(doc), &out)
assert.Error(t, err)
var derr *DecodeError
if !errors.As(err, &derr) {
t.Fatal("error not in expected format")
}
row, col := derr.Position()
assert.Equal(t, 2, row)
assert.Equal(t, 1, col)
assert.Equal(t, "toml: invalid character at start of key: =", derr.Error())
assert.Equal(t, `1| # comment
2| = "value"
| ~ invalid character at start of key: =`, derr.String())
}
func TestStrictErrorUnwrap(t *testing.T) {
fo := bytes.NewBufferString(`
Missing = 1
+12 -9
View File
@@ -704,15 +704,18 @@ func (enc *Encoder) encodeMap(b []byte, ctx encoderCtx, v reflect.Value) ([]byte
for iter.Next() {
v := iter.Value()
if isNil(v) {
// For nil pointers, convert to zero value of the element type.
// This allows round-trip marshaling of maps with nil pointer values.
// For nil interfaces and nil maps, skip since we can't derive a type.
if v.Kind() == reflect.Ptr {
// Handle nil values: convert nil pointers to zero value,
// skip nil interfaces and nil maps.
switch v.Kind() {
case reflect.Ptr:
if v.IsNil() {
v = reflect.Zero(v.Type().Elem())
} else {
}
case reflect.Interface, reflect.Map:
if v.IsNil() {
continue
}
default:
}
k, err := enc.keyToString(iter.Key())
@@ -936,7 +939,7 @@ func (enc *Encoder) encodeTable(b []byte, ctx encoderCtx, t table) ([]byte, erro
if shouldOmitEmpty(kv.Options, kv.Value) {
continue
}
if shouldOmitZero(kv.Options, kv.Value) {
if kv.Options.omitzero && shouldOmitZero(kv.Options, kv.Value) {
continue
}
hasNonEmptyKV = true
@@ -958,7 +961,7 @@ func (enc *Encoder) encodeTable(b []byte, ctx encoderCtx, t table) ([]byte, erro
if shouldOmitEmpty(table.Options, table.Value) {
continue
}
if shouldOmitZero(table.Options, table.Value) {
if table.Options.omitzero && shouldOmitZero(table.Options, table.Value) {
continue
}
if first {
@@ -995,7 +998,7 @@ func (enc *Encoder) encodeTableInline(b []byte, ctx encoderCtx, t table) ([]byte
if shouldOmitEmpty(kv.Options, kv.Value) {
continue
}
if shouldOmitZero(kv.Options, kv.Value) {
if kv.Options.omitzero && shouldOmitZero(kv.Options, kv.Value) {
continue
}
+7 -3
View File
@@ -28,12 +28,16 @@ func (c *Iterator) Next() bool {
if c.nodes == nil {
return false
}
nodes := *c.nodes
if !c.started {
c.started = true
} else if c.idx >= 0 {
c.idx = (*c.nodes)[c.idx].next
} else {
idx := c.idx
if idx >= 0 && int(idx) < len(nodes) {
c.idx = nodes[idx].next
}
}
return c.idx >= 0 && int(c.idx) < len(*c.nodes)
return c.idx >= 0 && int(c.idx) < len(nodes)
}
// IsLast returns true if the current node of the iterator is the last
+20 -6
View File
@@ -3,6 +3,7 @@ package unstable
import (
"bytes"
"fmt"
"reflect"
"unicode"
"github.com/pelletier/go-toml/v2/internal/characters"
@@ -83,10 +84,22 @@ func (p *Parser) rangeOfToken(token, rest []byte) Range {
}
// subsliceOffset returns the byte offset of subslice b within p.data.
// b must be a suffix (tail) of p.data.
// b must share the same backing array as p.data.
func (p *Parser) subsliceOffset(b []byte) int {
// b is a suffix of p.data, so its offset is len(p.data) - len(b)
return len(p.data) - len(b)
if len(b) == 0 {
// Most callers pass suffix slices, so preserve EOF behavior.
return len(p.data)
}
dataPtr := reflect.ValueOf(p.data).Pointer()
subPtr := reflect.ValueOf(b).Pointer()
offset := int(subPtr - dataPtr)
if offset < 0 || offset+len(b) > len(p.data) {
panic("subslice is not within parser input")
}
return offset
}
// Raw returns the slice corresponding to the bytes in the given range.
@@ -363,9 +376,10 @@ func (p *Parser) parseKeyval(b []byte) (reference, []byte, error) {
p.builder.Chain(valRef, key)
p.builder.AttachChild(ref, valRef)
// Set Raw to span the entire key-value expression
node := p.builder.NodeAt(ref)
node.Raw = p.rangeOfToken(startB[:len(startB)-len(b)], b)
// Set Raw to span the entire key-value expression.
// Access the node directly in the slice to avoid the write barrier
// that NodeAt's nodes-pointer setup would trigger.
p.builder.tree.nodes[ref].Raw = p.rangeOfToken(startB[:len(startB)-len(b)], b)
return ref, b, err
}