documentlanguagemodel Public

Watch 0 Fork 0 Star 0
markdown · 4349 bytes Raw Blame History

.dlm/ignore reference

A plain-text file that lives beside .dlm/training.yaml and carries one ignore pattern per line. Same anchor, different file — a repo can have either, both, or neither.

Use .dlm/ignore when you want drive-by exclusions without writing YAML. A three-line file is legal and common.

!!! info ".dlm (file) vs .dlm/ (directory) — two different things" Both .dlm shapes live in this project and it's worth naming the distinction once:

- `mydoc.dlm` — **a file** with a `.dlm` extension. The document
  itself, with YAML frontmatter + markdown body. Plural: users own
  many `mydoc.dlm` files, one per trainable corpus.
- `.dlm/` — **a directory** named literally `.dlm`. Lives inside a
  codebase or data tree and holds `training.yaml` + `ignore` +
  (when auto-scaffolded by `dlm train <dir>`) a `corpus.dlm` file.

The two share a name but not a namespace. When `dlm train <dir>`
auto-scaffolds, it creates a `.dlm/` directory inside `<dir>` and
places `corpus.dlm` inside it — so a single tree can end up with
`<dir>/.dlm/corpus.dlm`.

Example

# <repo-root>/.dlm/ignore
# Comments start with #. Blank lines are skipped.

*.min.js
**/fixtures/huge_*.json
docs/generated/

# Negate to re-include a file an earlier pattern excluded:
!docs/generated/README.md

# Trailing / = directory-only
build/

# Leading / = anchored to this .dlm/'s parent (not any ancestor)
/scripts/local-dev.sh

Grammar — supported

Syntax Meaning
# comment Line comment. Leading whitespace allowed before #.
blank line Skipped.
pattern Exclude match.
!pattern Re-include (negation) what an earlier pattern excluded.
pattern/ Directory-only — match only dirs, or paths under a dir that matches.
/pattern Anchored — match only at the .dlm/'s parent (not any ancestor).
** Globstar — matches zero or more path segments.
* Matches one path segment (non-/).
? Matches one character (non-/).

Grammar — NOT supported

The parser is a strict subset of .gitignore. Users coming from git may expect these and be surprised:

Missing feature Workaround
Character classes [abc] List patterns separately: a.py, b.py, c.py.
Backslash escapes Avoid literal *, ?, #, !, / at the start of a pattern; rename files if you must.
Whitespace-escape with backslash File names with trailing whitespace aren't supportable.

If you need any of the above, fall back to training.yaml's exclude: list — same glob grammar as include:, same fnmatch-style semantics as elsewhere in dlm.

Semantics — last-match-wins

Rules are evaluated top-to-bottom within a single file, and outer anchors first across the ancestor chain. For each candidate file, the last rule to match determines the verdict:

  • Match a non-negation rule → excluded.
  • Match a !negation rule → re-included.
  • No match → pass through (exclude/include resolution falls to the outer layers — parent directive, defaults, training.yaml).

This mirrors .gitignore exactly, so intuition transfers.

Worked example

Tree:

repo/
  .dlm/
    ignore           *.log\n!special.log\n
  debug.log
  special.log
  docs/
    .dlm/
      ignore         !*.log\n
    docs-only.log

Verdict (parent directive: include: ['**/*']):

File Outcome Why
repo/debug.log excluded Matches *.log, no later rule reopens it.
repo/special.log included *.log excludes, !special.log re-includes.
repo/docs/docs-only.log included Parent *.log excludes; docs/.dlm/ignore's !*.log re-includes for this subtree.

Error tolerance

Malformed lines log one WARN and are dropped — a typo in one line never fails the walk. Bare ! or bare / (no pattern after the sigil) are treated as malformed.

Relationship with .dlm/training.yaml

At the same .dlm/ anchor, both files coexist. Their exclude rules union. Because .dlm/ignore is evaluated after training.yaml.exclude at the same anchor, a .dlm/ignore !negation can re-include a file that training.yaml would otherwise drop. (Reason: .gitignore users expect ignore-file rules to be the final word at their anchor.)

View source
1 # `.dlm/ignore` reference
2
3 A plain-text file that lives beside `.dlm/training.yaml` and carries
4 one ignore pattern per line. Same anchor, different file — a repo
5 can have either, both, or neither.
6
7 Use `.dlm/ignore` when you want drive-by exclusions without writing
8 YAML. A three-line file is legal and common.
9
10 !!! info "`.dlm` (file) vs `.dlm/` (directory) — two different things"
11 Both `.dlm` shapes live in this project and it's worth naming the
12 distinction once:
13
14 - `mydoc.dlm` — **a file** with a `.dlm` extension. The document
15 itself, with YAML frontmatter + markdown body. Plural: users own
16 many `mydoc.dlm` files, one per trainable corpus.
17 - `.dlm/` — **a directory** named literally `.dlm`. Lives inside a
18 codebase or data tree and holds `training.yaml` + `ignore` +
19 (when auto-scaffolded by `dlm train <dir>`) a `corpus.dlm` file.
20
21 The two share a name but not a namespace. When `dlm train <dir>`
22 auto-scaffolds, it creates a `.dlm/` directory inside `<dir>` and
23 places `corpus.dlm` inside it — so a single tree can end up with
24 `<dir>/.dlm/corpus.dlm`.
25
26 ## Example
27
28 ```
29 # <repo-root>/.dlm/ignore
30 # Comments start with #. Blank lines are skipped.
31
32 *.min.js
33 **/fixtures/huge_*.json
34 docs/generated/
35
36 # Negate to re-include a file an earlier pattern excluded:
37 !docs/generated/README.md
38
39 # Trailing / = directory-only
40 build/
41
42 # Leading / = anchored to this .dlm/'s parent (not any ancestor)
43 /scripts/local-dev.sh
44 ```
45
46 ## Grammar — supported
47
48 | Syntax | Meaning |
49 |---|---|
50 | `# comment` | Line comment. Leading whitespace allowed before `#`. |
51 | blank line | Skipped. |
52 | `pattern` | Exclude match. |
53 | `!pattern` | Re-include (negation) what an earlier pattern excluded. |
54 | `pattern/` | Directory-only — match only dirs, or paths under a dir that matches. |
55 | `/pattern` | Anchored — match only at the `.dlm/`'s parent (not any ancestor). |
56 | `**` | Globstar — matches zero or more path segments. |
57 | `*` | Matches one path segment (non-`/`). |
58 | `?` | Matches one character (non-`/`). |
59
60 ## Grammar — NOT supported
61
62 The parser is a **strict subset** of `.gitignore`. Users coming from
63 git may expect these and be surprised:
64
65 | Missing feature | Workaround |
66 |---|---|
67 | Character classes `[abc]` | List patterns separately: `a.py`, `b.py`, `c.py`. |
68 | Backslash escapes | Avoid literal `*`, `?`, `#`, `!`, `/` at the start of a pattern; rename files if you must. |
69 | Whitespace-escape with backslash | File names with trailing whitespace aren't supportable. |
70
71 If you need any of the above, fall back to `training.yaml`'s
72 `exclude:` list — same glob grammar as `include:`, same fnmatch-style
73 semantics as elsewhere in dlm.
74
75 ## Semantics — last-match-wins
76
77 Rules are evaluated top-to-bottom within a single file, and outer
78 anchors first across the ancestor chain. For each candidate file,
79 the **last** rule to match determines the verdict:
80
81 - Match a non-negation rule → excluded.
82 - Match a `!negation` rule → re-included.
83 - No match → pass through (exclude/include resolution falls to the
84 outer layers — parent directive, defaults, `training.yaml`).
85
86 This mirrors `.gitignore` exactly, so intuition transfers.
87
88 ## Worked example
89
90 Tree:
91
92 ```
93 repo/
94 .dlm/
95 ignore *.log\n!special.log\n
96 debug.log
97 special.log
98 docs/
99 .dlm/
100 ignore !*.log\n
101 docs-only.log
102 ```
103
104 Verdict (parent directive: `include: ['**/*']`):
105
106 | File | Outcome | Why |
107 |---|---|---|
108 | `repo/debug.log` | excluded | Matches `*.log`, no later rule reopens it. |
109 | `repo/special.log` | included | `*.log` excludes, `!special.log` re-includes. |
110 | `repo/docs/docs-only.log` | included | Parent `*.log` excludes; `docs/.dlm/ignore`'s `!*.log` re-includes for this subtree. |
111
112 ## Error tolerance
113
114 Malformed lines log one WARN and are dropped — a typo in one line
115 never fails the walk. Bare `!` or bare `/` (no pattern after the
116 sigil) are treated as malformed.
117
118 ## Relationship with `.dlm/training.yaml`
119
120 At the same `.dlm/` anchor, both files coexist. Their exclude rules
121 union. Because `.dlm/ignore` is evaluated **after**
122 `training.yaml.exclude` at the same anchor, a `.dlm/ignore`
123 `!negation` can re-include a file that `training.yaml` would
124 otherwise drop. (Reason: `.gitignore` users expect ignore-file rules
125 to be the final word at their anchor.)