954a9ea · favela-bridge

Changed files (4) +132 -11

index.js +43 -2

…	…	@@ -21,7 +21,7 @@
21	21	import process from 'bare-process'
22	22	import os from 'bare-os'
23	23	import { join } from 'bare-path'
24		import { planRecord } from './lib/plan.js'
	24	import { planRecord, selectImages } from './lib/plan.js'
25	25	import { planProfile, sameProfile } from './lib/profile.js'
26	26	import { FavelaPeer } from './lib/peer.js'
27	27	import { BskySink } from './lib/bsky.js'
…	…	@@ -131,9 +131,13 @@ async function cycle (send) {
131	131	console.log(line + ' bsky: held — --limit ' + LIMIT + ' reached this cycle')
132	132	} else if (!send) {
133	133	console.log(line + ' bsky: would send ' + JSON.stringify(plan.bsky.record.text.slice(0, 48)) +
	134	(plan.bsky.images.length ? ' + ' + plan.bsky.images.length + ' image(s)' : '') +
134	135	(plan.bsky.losses.length ? ' — losing: ' + plan.bsky.losses.join('; ') : ''))
135	136	} else {
136		const r = await bsky.send(row.name, plan.bsky.record)
	137	const { embed, losses: imgLosses } = await buildEmbed(peer, bsky, plan.bsky.images)
	138	for (const l of imgLosses) console.log(line + ' bsky: ' + l)
	139	const record = embed ? { ...plan.bsky.record, embed } : plan.bsky.record
	140	const r = await bsky.send(row.name, record)
137	141	if (r.sent) { sent++; console.log(line + ' bsky: SENT ' + r.rkey + (plan.bsky.losses.length ? ' — lost: ' + plan.bsky.losses.join('; ') : '')) }
138	142	else console.log(line + ' bsky: FAILED — ' + r.error)
139	143	}
…	…	@@ -160,6 +164,43 @@ async function cycle (send) {
160	164	*/
161	165	const blobCache = new Map() // oid -> blob ref. Content-addressed, so this can never go stale.
162	166
	167	/**
	168	* Turn the oids planBsky named into an `app.bsky.embed.images`, uploading what is missing.
	169	*
	170	* ⛔ THE PICTURE IS WHY THE POST EXISTS, quite often — and until now every one of them reached the
	171	* fediverse and none reached bsky, reported as "bsky needs an uploaded blob ref, which is unbuilt".
	172	* It was unbuilt because nothing here could upload; hpp has always had upload-blob.
	173	*
	174	* ⚠ A post whose images ALL fail still goes, as text. The alternative is withholding a post because
	175	* a picture would not upload, which loses more than it protects.
	176	*/
	177	async function buildEmbed (peer, bsky, imgs) {
	178	if (!imgs \|\| !imgs.length) return { embed: null, losses: [] }
	179
	180	const described = []
	181	for (const im of imgs) {
	182	let f = null
	183	try { f = await peer.call('file-get', { oid: im.oid }) } catch { f = null }
	184	described.push({ ...im, path: (f && f.path) \|\| null, mimeType: f && f.mimeType, size: f && f.size })
	185	}
	186
	187	const { images, losses } = selectImages(described)
	188	const out = []
	189	for (const d of images) {
	190	let ref = blobCache.get(d.oid)
	191	if (!ref) {
	192	ref = await bsky.uploadBlob(d.path, d.mimeType)
	193	if (ref) blobCache.set(d.oid, ref)
	194	}
	195	// ⛔ Alt text travels. It is an accessibility obligation and the author already wrote it.
	196	if (ref) out.push({ alt: d.alt \|\| '', image: ref })
	197	else losses.push('an image could not be uploaded — omitted from this post')
	198	}
	199
	200	if (!out.length) return { embed: null, losses }
	201	return { embed: { $type: 'app.bsky.embed.images', images: out }, losses }
	202	}
	203
163	204	async function reconcileProfile (peer, bsky, reach, send) {
164	205	if (!bsky.configured) return
165	206	// ⛔ CONSENT FIRST, exactly as for a post. A profile is published content too.

lib/plan.js +52 -5

…	…	@@ -23,7 +23,7 @@
23	23
24	24	import { maxReach } from 'favela-markdown/lib/reach.js'
25	25	import { strip } from 'favela-markdown/lib/strip.js'
26		import { render } from 'favela-markdown/lib/render.js'
	26	import { render, embeds } from 'favela-markdown/lib/render.js'
27	27
28	28	// app.bsky.feed.post caps `text` at 300 GRAPHEMES and 3000 bytes. Graphemes, not characters — an
29	29	// emoji is one grapheme and four bytes, and counting the wrong one rejects valid posts.
…	…	@@ -32,6 +32,37 @@ const MAX_BYTES = 3000
32	32
33	33	const BLOB_RE = /!\[[^\]]*\]\(favela:\/\/blob\//
34	34
	35	// app.bsky.embed.images: at most 4, and each blob at most 1,000,000 bytes.
	36	export const MAX_IMAGES = 4
	37	export const MAX_IMAGE_BYTES = 1000000
	38
	39	/**
	40	* Which described images may actually be sent. Pure.
	41	*
	42	* ⛔ WITHHELD, NEVER RESIZED. We do not re-encode anyone's picture: favela's own cap is 512 KiB for
	43	* an avatar but a POST attachment has no cap at all (the overlay chunks), so a photo can and will
	44	* exceed what bsky takes. Shrinking it would publish a degraded copy under the author's name with
	45	* nothing to say it had been altered — the same argument that makes plan.js withhold a long post
	46	* rather than cut it. The record stays whole on favela and the loss is named.
	47	*/
	48	export function selectImages (described) {
	49	const images = []
	50	const losses = []
	51	for (const d of described \|\| []) {
	52	if (!d \|\| !d.path) { losses.push('an image could not be read from the peer'); continue }
	53	if (!/^image\//.test(d.mimeType \|\| '')) {
	54	losses.push('image omitted — bsky takes image/*, this is ' + (d.mimeType \|\| 'untyped'))
	55	continue
	56	}
	57	if (typeof d.size === 'number' && d.size > MAX_IMAGE_BYTES) {
	58	losses.push('image omitted — ' + d.size + ' bytes against bsky\'s limit of ' + MAX_IMAGE_BYTES + ' (not resized: that would publish an altered copy)')
	59	continue
	60	}
	61	images.push(d)
	62	}
	63	return { images, losses }
	64	}
	65
35	66	const byteLength = (s) => (typeof Buffer !== 'undefined' && Buffer.byteLength)
36	67	? Buffer.byteLength(s, 'utf8')
37	68	: new TextEncoder().encode(s).length
…	…	@@ -91,10 +122,24 @@ function planBsky (body, createdAt) {
91	122	return WITHHELD('could not strip the record: ' + ((err && err.message) \|\| String(err)))
92	123	}
93	124
94		// ⚠ Counted from the SOURCE. strip() removes an image entirely, so by the time we hold `text`
95		// there is no evidence a picture was ever in the post.
96		if (BLOB_RE.test(body)) {
97		losses.push('attachment(s) not projected — bsky needs an uploaded blob ref, which is unbuilt')
	125	// ⚠ Named from the SOURCE. strip() removes an image entirely, so by the time we hold `text` there
	126	// is no evidence a picture was ever in the post — which is why the oids are collected here, off
	127	// the same parse render uses, and never off a regex over the stripped text.
	128	//
	129	// ⛔ NAMED, NOT UPLOADED. Resolving an oid to a blob ref is I/O; this module is pure. The caller
	130	// hands back refs and builds the embed, exactly as the profile projection does.
	131	let images = []
	132	try {
	133	images = (embeds(body) \|\| []).filter((e) => e && e.oid && e.kind === 'image').map((e) => ({ oid: e.oid, alt: e.alt \|\| '' }))
	134	} catch { images = [] }
	135	if (!images.length && BLOB_RE.test(body)) {
	136	// The regex sees a blob embed the parser did not return as an image — a link to a file rather
	137	// than a picture. Reported so "no images" and "a file bsky has no form for" stay distinguishable.
	138	losses.push('attachment(s) not projected — the embed is not an image')
	139	}
	140	if (images.length > MAX_IMAGES) {
	141	losses.push('only the first ' + MAX_IMAGES + ' of ' + images.length + ' images projected — bsky takes ' + MAX_IMAGES)
	142	images = images.slice(0, MAX_IMAGES)
98	143	}
99	144
100	145	if (!text.trim()) return { project: false, losses: losses.concat(['nothing to post — no text survives the strip']) }
…	…	@@ -119,6 +164,8 @@ function planBsky (body, createdAt) {
119	164	// ⚠ The AUTHOR's time, unchanged. Stamping projection time would make every backfilled record
120	165	// claim to be new; the field is the author's statement about when they wrote it.
121	166	record: { $type: 'app.bsky.feed.post', text, createdAt },
	167	// The caller uploads these and attaches the embed; an empty list means there was nothing to send.
	168	images,
122	169	losses
123	170	}
124	171	}

package.json +1 -1

…	…	@@ -1,6 +1,6 @@
1	1	{
2	2	"name": "favela-bridge",
3		"version": "0.0.9",
	3	"version": "0.0.10",
4	4	"description": "Projects favela records onto the networks a room may reach. One reader, one reach decision, one projection per network.",
5	5	"type": "module",
6	6	"main": "index.js",

test/plan.test.js +36 -3

…	…	@@ -13,7 +13,7 @@
13	13	// projection loses what the second one never had.
14	14
15	15	import test from 'brittle'
16		import { planRecord } from '../lib/plan.js'
	16	import { planRecord, selectImages, MAX_IMAGE_BYTES } from '../lib/plan.js'
17	17
18	18	const at = '2026-09-07T23:12:04.600Z'
19	19	const row = (body, extra = {}) => ({ name: '1788822724600', body, createdAt: at, ...extra })
…	…	@@ -69,8 +69,41 @@ test('⛔ too long for bsky is WITHHELD there and still sent to the fediverse',
69	69	})
70	70
71	71	test('⛔ losses are reported on a SUCCESSFUL projection too', async (t) => {
72		const p = planRecord(row('look ![p.webp](favela://blob/' + 'cd'.repeat(32) + ')'), { roomType: 'public', verb: 'post' })
73		t.ok(p.bsky.losses.length > 0, 'the picture is named: ' + p.bsky.losses.join('; '))
	72	// A mention still cannot be projected — resolving favela:// to a did:plc: is unbuilt — and the
	73	// post goes anyway with the loss named beside it.
	74	const p = planRecord(row('hello [@alice](favela://' + 'ab'.repeat(32) + ')'), { roomType: 'public', verb: 'post' })
	75	t.ok(p.bsky.project, 'the post still goes')
	76	t.ok(p.bsky.losses.length > 0, 'and the loss is named: ' + p.bsky.losses.join('; '))
	77	})
	78
	79	// ⛔ THIS USED TO BE A LOSS. "attachment(s) not projected — bsky needs an uploaded blob ref, which
	80	// is unbuilt" was reported for every post carrying a picture, and the picture reached the fediverse
	81	// and never bsky. The oids are named here; the caller uploads them and attaches the embed.
	82	test('★ a post image is NAMED for projection rather than reported as a loss', async (t) => {
	83	const oid = 'cd'.repeat(32)
	84	const p = planRecord(row('look ![a cat](favela://blob/' + oid + ')'), { roomType: 'public', verb: 'post' })
	85	t.is(p.bsky.images.length, 1, 'the picture is offered for upload')
	86	t.is(p.bsky.images[0].oid, oid)
	87	t.is(p.bsky.images[0].alt, 'a cat', '⛔ alt text is an accessibility obligation, not a nicety')
	88	t.absent(p.bsky.losses.some((l) => l.includes('unbuilt')), 'and is no longer reported as unbuilt')
	89	})
	90
	91	test('at most four images, and the rest are named as dropped', async (t) => {
	92	const imgs = Array.from({ length: 6 }, (_, i) => '![' + i + '](favela://blob/' + String(i).repeat(64) + ')').join(' ')
	93	const p = planRecord(row('many ' + imgs), { roomType: 'public', verb: 'post' })
	94	t.is(p.bsky.images.length, 4, 'bsky takes four')
	95	t.ok(p.bsky.losses.some((l) => l.includes('only the first 4 of 6')), 'and says so')
	96	})
	97
	98	test('⛔ selectImages WITHHOLDS an oversized picture rather than resizing it', async (t) => {
	99	const { images, losses } = selectImages([
	100	{ oid: 'a', path: '/tmp/a.png', mimeType: 'image/png', size: 500 },
	101	{ oid: 'b', path: '/tmp/b.png', mimeType: 'image/png', size: MAX_IMAGE_BYTES + 1 },
	102	{ oid: 'c', path: '/tmp/c.pdf', mimeType: 'application/pdf', size: 10 }
	103	])
	104	t.is(images.length, 1, 'only the one bsky will take')
	105	t.ok(losses.some((l) => l.includes('against bsky')), 'the oversized one is named, not shrunk')
	106	t.ok(losses.some((l) => l.includes('image/*')), 'and so is the non-image')
74	107	})
75	108
76	109	test('⛔ planRecord NEVER throws — it runs on the projection path', async (t) => {

favela-bridge commits / 954a9ea

Changed files (4) +132 -11