1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
//! The nomination gate: which branch-relative paths a compaction pass
//! may **not** shed (ARCH §2.7).
//!
//! **What is not the branch's history is not a pass's to shed** (§2.7,
//! §2.8; bl-898f, bl-541b, bl-c7bb), and
//! [`not_compaction_eligible`] is the one predicate saying which paths
//! those are. Three classes are in it.
//!
//! The **dispatch entry** is the goal in transcript form. A goal has two
//! projections written at one dispatch and neither ever rewritten — the
//! pinned `goal.md` (§2.8) and the entry the same text was deposited as
//! through the front door (§2.11: a root's opening user message, a
//! child's dispatch message). The compactor is shown one of them — its
//! own goal quotes the dispatching branch's `goal.md` verbatim (§2.7) —
//! while the other sits in the transcript it is told to prune, so that
//! entry is the one that reads as *pure duplication* to a model
//! nominating superseded files. It was nominated and deleted in
//! practice, and what it deleted was the operator's only copy of the
//! prompt the conversation exists to serve.
//!
//! The **dispatch-written head** is the second class, and strictly
//! worse when it fires: the system slot's files — `goal.md`, `soul.md`
//! and `name` ([`crate::prompt::dispatch::step_commit::SYSTEM_SLOT_FILES`],
//! §5.2 structural wire homes) — and the lineage's `facts.md`
//! ([`crate::facts`], §5.5), cut into the tree at the same dispatch
//! commit and pinned at the head of every call after it. A compactor
//! writes its own at its dispatch commit, so a nomination of one after
//! that is a deletion inside the `dispatch..tip` range the landing
//! classifies as the compactor's product (`compactor::land`): it lands
//! as a `git rm` against the *dispatching* branch's tree, which then
//! keeps stepping with no goal, no soul, no identity line or no durable
//! memory. One class and not two: a dispatch-written fact is not the
//! branch's history to shed.
//!
//! **This pass's own product** is the third, and the one with the
//! largest blast radius (bl-c7bb). A live compactor accepted
//! `mark_for_deletion` of the `summary/001.md` it had written seconds
//! earlier through the other half of this same pair; the landing admits
//! the summary and the deletions and nothing else (§2.6), so a landing
//! carrying a `git rm` of its own summary carries away the entire span
//! it was dispatched to preserve and leaves a base holding nothing of
//! it. A **blanket** `summary/**` refusal would be wrong — superseding
//! an *earlier* pass's summary is the shipped soul's own instruction
//! (`template/souls/compactor.md`) and the only thing that keeps the
//! chain from growing without bound — so the class is *this run's*
//! output, not the directory. It is read from the same place the landing
//! reads the compaction product from ([`super::land::base`]): what
//! changed after the compactor's **own dispatch commit** is the pass's,
//! and everything at or before it is the inherited branch. One
//! definition of "the product", now read by the landing that carries it
//! and by the door that refuses to shed it.
//!
//! **The landing's own product needs no fourth class.** Since bl-e655 a
//! landing writes an **extract** beside the summary
//! (`summary/NNN.refs.md`, [`super::land::extract`]), and ARCH §2.7 puts
//! *this pass's* extract in the not-eligible class exactly as this pass's
//! summary. It is there by the same predicate and by construction: the
//! extract is derived and staged by the landing, into the *dispatching*
//! branch's base commit, so it never exists on the compactor's branch for
//! the pass to nominate. An **earlier** pass's extract does exist in the
//! inherited tree, at or before this compactor's dispatch commit, and is
//! nominable — the same distinction its summary draws, for the same
//! reason: superseding it is the compaction.
//!
//! All three are declined **at the nomination**, in-band, so the
//! compactor's summary is never premised on a deletion that did not
//! happen. Live-branch-wins is dropped at the landing instead, precisely
//! because *its* fact — a race with the live branch — is not knowable
//! when the compactor nominates (§2.6).
use Error;
use crateMESSAGES_DIR;
use crateGitRunner;
use Path;
/// Transcript counter of the **dispatch entry** — the entry every
/// branch's opening prompt lands as (module docs, §2.3, §2.11). The
/// counter is monotonic and never reused (`dispatch::transcript`), so
/// `001` names that entry for the life of the branch.
const DISPATCH_SEQ: u32 = 1;
/// What makes the branch-relative `path` un-eligible for compaction
/// (module docs), or `None` when nothing does. The returned phrase is
/// the noun the decline names it by, so the model is told which rule it
/// hit rather than being refused anonymously.
///
/// The first two classes are derived from the path alone — the dispatch
/// entry from the same `NNN-` prefix reading `dispatch::transcript`'s
/// counter uses, the system slot from the file names the composer pins —
/// so they need no tree, no config and no state, and they hold for a
/// `.md` delivery and a resumed conversation's inherited first entry
/// alike. The third cannot be: "whose output is this file" is not in the
/// name, and a blanket `summary/**` refusal would forbid the supersede
/// the chain depends on (module docs). It reads the branch instead, at
/// the one anchor the landing already classifies the product by.
pub
/// Has *this* compaction pass written `rel` — is it an addition or a
/// rewrite in the range after the compactor's own dispatch commit?
///
/// That range is the landing's definition of the compaction product
/// ([`super::land::base`]: "a path *added under `summary/`* is the
/// `write_summary` product"), read here against the **index** rather
/// than a commit pair, which is exactly the set `git rm` could carry
/// away: a summary already committed by its tool step and one merely
/// staged both answer the same, and an untracked one answers `false`
/// because `git rm` refuses it anyway — the nonexistent-path decline
/// takes that case, and nothing is lost either way.
///
/// `--diff-filter=AM` excludes deletions on purpose: a path this pass
/// already staged for removal is not its product, and a re-nomination of
/// one should meet the nonexistent-path decline it has always met, not
/// this one.
///
/// No dispatch commit reachable — a tree that was never founded — is the
/// general path with empty inputs: nothing was added after a commit that
/// does not exist, so nothing is this pass's ([`crate::prompt::role::founding_sha`]
/// answers `None` the same way [`super::checkpoint::origin`] does).