TOML encoder: prefer readable table sections over inline tables

[Copilot]

When encoding to TOML from YAML/JSON, nested mappings were collapsed into compact inline table syntax (key = { ... }) instead of idiomatic TOML table sections ([key]). This affected both direct YAML→TOML conversion and round-trips through YAML.

Changes

Encoder (encoder_toml.go)

• encodeTopLevelEntry: Check EncodeHintInline/FlowStyle — only nodes marked as inline stay inline; all block mappings become table sections
• encodeSeparateMapping: Count EncodeHintInline/FlowStyle mapping children as hasAttrs (they render as key = { ... }); use recursive encodeSeparateMapping in the "no attrs" loop
• encodeMappingBodyWithPath: Emit non-inline child mappings as sub-table sections ([parent.child]) instead of inline tables

Decoder (decoder_toml.go)

• Set EncodeHint: EncodeHintInline on decoded TOML inline tables, distinguishing them from block mappings so they survive re-encoding as inline
• Set EncodeHint: EncodeHintSeparateBlock on decoded TOML table sections ([section]) and array tables ([[section]])

Core (candidate_node.go)

• Replaced EncodeSeparate bool and TomlInline bool fields with a single EncodeHint EncodeHint enum field, eliminating the possibility of conflicting boolean combinations and making it easy to add further encoding options in future:
  • EncodeHintDefault (0) — encoder picks the representation (TOML block mappings default to [section])
  • EncodeHintSeparateBlock — force emission as a separate block/section header (TOML [section], HCL blocks)
  • EncodeHintInline — force emission as an inline/flow table (TOML inline tables)
• UpdateAttributesFrom propagates EncodeHint so it survives DeeplyAssign deep-merges

Decoder/Encoder HCL (decoder_hcl.go, encoder_hcl.go)

• Updated to use EncodeHint: EncodeHintSeparateBlock in place of the old EncodeSeparate: true

Tests (toml_test.go)

• Add encode scenario type (YAML→TOML) and coverage for table section output
• TOML roundtrip tests preserved: inline tables stay inline, table sections stay as sections

Before / After

# input YAML
arg:
  hello: foo

# before
arg = { hello = "foo" }

# after
[arg]
hello = "foo"

YAML flow mappings (arg: {hello: foo}) and TOML inline tables are preserved as TOML inline tables, respecting the author's explicit style intent. TOML table sections and YAML block mappings both produce readable [section] headers.

[mikefarah]

@copilot the point of the roundtrip tests is to maintain the format as much as possible...this no longer does that.

[Copilot]

Fixed in d1ffec1. The root cause was that FlowStyle (used in the previous commit to mark inline tables) also affects the YAML encoder, which was changing the TOML→YAML decode output.

The fix introduces a new TomlInline bool field on CandidateNode — TOML-specific and invisible to the YAML encoder. The TOML decoder sets it for inline tables, UpdateAttributesFrom propagates it through DeeplyAssign merges, and the TOML encoder checks it to decide inline vs table-section output.

TOML roundtrip tests are restored to their original expected values: inline tables stay inline, table sections stay as table sections.

[mikefarah]

@copilot how is TomlInline different to EncodeSeparate? They seem quite similar right? Perhaps EncodeSeparete just needs to default to true when decoding toml or something? Rather than adding another new property that's specific to TOML

[Copilot]

They're semantically opposite:

• EncodeSeparate = true → "render as a [section] header"
• TomlInline = true → "render as key = { ... } inline table"

Three cases need to be distinguished:

Source	EncodeSeparate	TomlInline	Output
TOML table section	true	false	[section]
TOML inline table	false	true	key = { }
YAML block mapping	false	false	[section] (encoder default)

Without TomlInline, cases 2 and 3 look identical (EncodeSeparate = false) so inline tables would expand to sections in roundtrip.

FlowStyle was the approach in the previous commit (b99f417), but it has two problems: (1) its propagation in UpdateAttributesFrom is conditional (if n.Style == 0), so it gets lost during DeeplyAssign's deep-merge — confirmed empirically, which is exactly why b99f417 broke roundtrip; (2) it changes TOML→YAML decode output (inline tables become flow YAML instead of block YAML).

TomlInline uses the same unconditional propagation pattern as EncodeSeparate, which is why it works reliably. Happy to explore any alternative approach if you have one in mind.

[mikefarah]

@copilot

What about EncodeSeparate = true and TomlInline = true? That would be a conflict right?

I think booleans are going to get complex and conflict with each other. Maybe a better approach is to rethink EncodeSeparate, change it from a boolean to a int with bit flags EncodingOptions enumeration?

To being with, we could have 3 values:
Default (nil?): (encoder default) for TOML this would output "[section]"
SEPARATE_HEADER: explicitly output "[section]"
INLINE: explicitly render the table inline.

This lets us add more options later.

There's still a problem of specifying to conflicting bits though (e.g. SEPARATE_HEADER and INLINE). Not sure if there's something we can do that better..

[Copilot]

Done in 8c1cb6a. Replaced both EncodeSeparate bool and TomlInline bool with a single EncodeHint enum on CandidateNode:

type EncodeHint int

const (
    EncodeHintDefault      EncodeHint = iota // encoder default (TOML → [section])
    EncodeHintSeparateBlock                   // force separate block/section header
    EncodeHintInline                          // force inline / flow table
)

• TOML [section] decoder → EncodeHintSeparateBlock
• TOML inline table decoder → EncodeHintInline
• HCL multi-block decoder → EncodeHintSeparateBlock
• All other nodes → EncodeHintDefault (zero value, no change)

Adding more encoding options in the future is just a new constant; conflicting combinations are impossible since it's a single enum field.

[mikefarah]

@copilot fix the conflicts

[Copilot]

@copilot fix the conflicts

Done in a297c15. The only conflict was in toml_test.go — both branches added new test cases at the same location. Resolved by keeping both: the encode-scenario tests from this PR and the special-character roundtrip tests from master. All tests pass.

[mikefarah]

@copilot Error: pkg/yqlib/encoder_toml.go:536:24: func (*tomlEncoder).hasStructuralChildren is unused (unused)

[Copilot]

@copilot Error: pkg/yqlib/encoder_toml.go:536:24: func (*tomlEncoder).hasStructuralChildren is unused (unused)

Fixed in 320b0de — removed the unused hasStructuralChildren method.