mcp-databnf 1.0.0 → 1.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +169 -0
- package/LICENSE +0 -0
- package/README.md +78 -18
- package/dist/bnf/client.d.ts +73 -0
- package/dist/bnf/client.js +207 -38
- package/dist/bnf/client.js.map +1 -1
- package/dist/index.d.ts +0 -0
- package/dist/index.js +728 -151
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -2,6 +2,175 @@
|
|
|
2
2
|
|
|
3
3
|
This project follows [semantic versioning](https://semver.org/).
|
|
4
4
|
|
|
5
|
+
## 1.1.0
|
|
6
|
+
|
|
7
|
+
- A text search reads a fixed window of the index, and a search that filled it
|
|
8
|
+
reported exhaustion. `search_authors {"name":"Marie","limit":50,"page":8}`
|
|
9
|
+
returned fifty rows and `has_more: false`, while page 9 was refused for
|
|
10
|
+
reaching past the end of what one search reads: of the two statements, the one
|
|
11
|
+
a caller reaches by stopping where the server told them to was the false one.
|
|
12
|
+
The window is also read before the type filter, so a list of a hundred and
|
|
13
|
+
fifty rows could rest on a full one. The endpoint now sends the occupancy of
|
|
14
|
+
the window in the request that reads the page, and both searches publish
|
|
15
|
+
`index_window_full` and say in words what a full window means: the rows are a
|
|
16
|
+
part of the matches, and `has_more` false marks where the reading stopped. No
|
|
17
|
+
count of matches is reported, here or anywhere else.
|
|
18
|
+
- The rows a full window holds are settled by the index as it answers, so two
|
|
19
|
+
readings of one search could differ with nothing said. The same wording now
|
|
20
|
+
states it, and ties it to the window: an answer resting on a window with room
|
|
21
|
+
to spare holds every match and is repeatable.
|
|
22
|
+
- A page emptied by the filter behind a full window said that the BnF held
|
|
23
|
+
nobody, and no reading had established that. Both searches now name the
|
|
24
|
+
window as what the page rests on and leave the question open.
|
|
25
|
+
- `words_searched` named one term where the index had required two. It cuts a
|
|
26
|
+
word at an apostrophe and at a hyphen, so `O'Brien` is required as `O` and
|
|
27
|
+
`Brien` and a record written `Pat O Brien` matches. Both searches now report
|
|
28
|
+
the terms the index required, and say which word was cut, so a caller can see
|
|
29
|
+
from the fields why a row is on the list.
|
|
30
|
+
- `get_author` said a person might be living on a record that dates them
|
|
31
|
+
nowhere, about a man dead since 1881. The BnF keeps a second authority record
|
|
32
|
+
for some people, carrying a name and nothing else, and a silence on that
|
|
33
|
+
record is a fact about the record. The note now turns on whether the record
|
|
34
|
+
states a birth, and a record stating neither date says so and points at
|
|
35
|
+
`search_authors` as the way to the other records the file holds.
|
|
36
|
+
- `list_works` answering with no row now carries the same pointer, since the
|
|
37
|
+
link it reports hangs off a record rather than off a person.
|
|
38
|
+
- `get_work` printed the work-form codes with none of the caveat the listing
|
|
39
|
+
gives them. The wording is now shared, so both tools say that the vocabulary
|
|
40
|
+
publishes no label for those terms and that an absent code is a form the
|
|
41
|
+
catalogue does not state.
|
|
42
|
+
- `search_authors`, `search_works`, `list_editions` and `list_works` served one
|
|
43
|
+
record on two pages and left another out. Each cuts its page in a subquery and
|
|
44
|
+
then reads the columns in an outer query that asked for no order, and SPARQL
|
|
45
|
+
gives such a query no order at all: the endpoint sent the rows of a page in a
|
|
46
|
+
sequence of its own, and the row dropped to turn a page of eleven into a page
|
|
47
|
+
of ten was whichever one arrived last. Marguerite Duras at four works a page
|
|
48
|
+
put `cb12285962p` on pages 2 and 3 and never showed `cb122859600`. Every paged
|
|
49
|
+
query now sorts its outer form on the same address the page was cut along, so
|
|
50
|
+
a walk through the pages reaches each record once. The sort runs over the rows
|
|
51
|
+
of a single page, and the endpoint answers it in the same time as before.
|
|
52
|
+
- The note attached to a list of works claimed the order held between pages,
|
|
53
|
+
which was the guarantee the query had not asked for. Every wording about the
|
|
54
|
+
order now names what the rows are ordered by, and says that paging along it
|
|
55
|
+
reaches each record once.
|
|
56
|
+
|
|
57
|
+
- `list_works` walks from a person to the works the catalogue names them the
|
|
58
|
+
creator of, which is the question "what did this person write". A title search
|
|
59
|
+
cannot answer it: the index reads titles, so a person's name in a title belongs
|
|
60
|
+
to the books written about them, and searching "Balzac" returns his critics.
|
|
61
|
+
- The list states what it is. It reports one link, `dcterms:creator`, and the
|
|
62
|
+
catalogue credits a person on a record in other ways and holds editions whose
|
|
63
|
+
work it has never established, so an absent title is not a title the person did
|
|
64
|
+
not write. No total is reported: a count of what one link reaches would be read
|
|
65
|
+
as a count of what somebody wrote.
|
|
66
|
+
- Each row carries the form codes the catalogue points at, untranslated, because
|
|
67
|
+
this dataset publishes no label for that vocabulary.
|
|
68
|
+
- A page holding no row is read before anything is said about the catalogue. A
|
|
69
|
+
page past the last row points back to page 1, a person the link is empty on is
|
|
70
|
+
named as such, and an address the catalogue types as something other than a
|
|
71
|
+
person is refused with what it is.
|
|
72
|
+
|
|
73
|
+
- `list_editions` answered a person's identifier with an empty list of editions
|
|
74
|
+
and explained it as a property of work records, which invented a bibliographic
|
|
75
|
+
fact about a record that is not a work. It now reads what the address is
|
|
76
|
+
before explaining anything, and refuses one the catalogue types otherwise,
|
|
77
|
+
naming that type and pointing at `list_works` and `get_author`. A provisional
|
|
78
|
+
identifier still costs no query, since only a work is addressed under
|
|
79
|
+
`temp-work`.
|
|
80
|
+
- `find_digitised` followed the `kind` it was given without checking it, and
|
|
81
|
+
returned that same word as its own reading of the record. The person path and
|
|
82
|
+
the work path reach different links, so a wrong kind answered with part of the
|
|
83
|
+
record and nothing said so. The catalogue is now asked what the record is on
|
|
84
|
+
every call, `kind` in the answer is that reading, and a stated kind the
|
|
85
|
+
catalogue contradicts is refused as invalid input.
|
|
86
|
+
- The wording about work forms denied information the field carries. Some terms
|
|
87
|
+
of the vocabulary read as words and some do not, and what the catalogue
|
|
88
|
+
withholds is a form on the many works that state none. Every place that
|
|
89
|
+
describes them, the tool descriptions, the notes, the schemas and the README,
|
|
90
|
+
now says that: a code is what a work declares, an absent code is a form
|
|
91
|
+
unstated, and selecting on a code finds the works declaring it and never all
|
|
92
|
+
the works of that form.
|
|
93
|
+
- A person's record can state its dates twice, in the brackets of a heading and
|
|
94
|
+
in the dated fields, and the two can disagree. `get_author` and
|
|
95
|
+
`search_authors` now say when they do, quoting both, and settle nothing: those
|
|
96
|
+
fields are what tells two people of one name apart, and the record gives no
|
|
97
|
+
ground for preferring either side.
|
|
98
|
+
- `search_authors` matches the index letter for letter, so a name written with
|
|
99
|
+
other accents or under another transliteration is a different search reaching
|
|
100
|
+
different records, and a short list of rows looks exactly like a complete one.
|
|
101
|
+
Every answer now says so. An answer holding no row also said that no record in
|
|
102
|
+
the authority file carried the name, having read the person records alone; it
|
|
103
|
+
now states that scope, so an organisation or a conference is named as outside
|
|
104
|
+
what was searched.
|
|
105
|
+
- A character that marks nothing on a screen was treated as a word separator, so
|
|
106
|
+
a control character pasted inside a name cut it in two and the index was asked
|
|
107
|
+
for both halves. `Rimbaud` carrying one such byte returned no row and the
|
|
108
|
+
answer read "No person in the BnF authority file matches", which no reading had
|
|
109
|
+
established. The control characters, the zero-width spaces and joiners, the
|
|
110
|
+
soft hyphen, the bidirectional marks and the byte order mark are now removed
|
|
111
|
+
before the words are cut, so a word one of them sits inside is searched for
|
|
112
|
+
whole, and the answer says how many were removed. A space, a tab and a line
|
|
113
|
+
break still separate words.
|
|
114
|
+
- Punctuation reached no term while a one-letter fragment was kept as a term the
|
|
115
|
+
index required, and neither was said. Both searches now name the characters
|
|
116
|
+
they set aside beside the terms they asked for, and name a term of one
|
|
117
|
+
character as required like any other, since either can be what emptied a list.
|
|
118
|
+
- The addresses under `same_as` arrived with nothing said about where they came
|
|
119
|
+
from or what is known of them. `get_author` and `get_work` now state that they
|
|
120
|
+
are the BnF's own alignments, carried character for character, and that this
|
|
121
|
+
server requests data.bnf.fr alone and does not open them, so whether one still
|
|
122
|
+
answers is something it has not looked at. Re-spelling one, by decoding an
|
|
123
|
+
escape or adding an escape of this server's own, would produce an address the
|
|
124
|
+
catalogue never published.
|
|
125
|
+
- `find_digitised` published its per-role figures under `counts`, next to
|
|
126
|
+
`has_more`, where they read as a count of what the catalogue attaches. They
|
|
127
|
+
count the links in the answer, and Baudelaire's record returns a hundred and
|
|
128
|
+
one of them with more left out at the ceiling. The field is now
|
|
129
|
+
`links_returned_by_role`, it says what it counts, and a shortened answer says
|
|
130
|
+
that the figures fall short by however much was left out.
|
|
131
|
+
- A digitised link carried an address opening a small image beside an ARK naming
|
|
132
|
+
the document that image was taken from, with `url` described as the document.
|
|
133
|
+
Sixty-three of the hundred and one links on Baudelaire's record are such
|
|
134
|
+
addresses. Every link now carries `rendering`, the view the address asks
|
|
135
|
+
Gallica for, read off what follows the ARK name and null when the address
|
|
136
|
+
names the document; the answer counts the links carrying one.
|
|
137
|
+
- A record stating one text under `biographical_information` and under
|
|
138
|
+
`occupation` was returned as two fields, where it reads as two statements
|
|
139
|
+
agreeing. `get_author` now says when the two are identical character for
|
|
140
|
+
character.
|
|
141
|
+
- One identifier was refused with a different list of types depending on which
|
|
142
|
+
tool refused it, because a detail query gathers the classes of everything it
|
|
143
|
+
walks through. Every refusal now states the list the address itself answers
|
|
144
|
+
with, so the five tools describe one record the same way.
|
|
145
|
+
- `catalogue_url` came back null on `get_work` while `get_author` and
|
|
146
|
+
`list_editions` filled it, with nothing to read that silence by. Both fields
|
|
147
|
+
now say that the address is the one the record points at, that null is the
|
|
148
|
+
record stating none, and that no address is built in its place.
|
|
149
|
+
- The square brackets of cataloguing passed through `publisher` and `place`
|
|
150
|
+
unexplained, so `[S.l.]`, `[s.n.]` and a role such as `[éd., distrib.]` read as
|
|
151
|
+
what the item prints. The values are still passed on as published, and the
|
|
152
|
+
fields and a note now say what the brackets mean.
|
|
153
|
+
- The Gallica caveat was attached to answers from `get_author` and `get_work`
|
|
154
|
+
that carried no link, where it describes a list that is not there. It now
|
|
155
|
+
travels with the images only when they are returned.
|
|
156
|
+
|
|
157
|
+
## 1.0.1
|
|
158
|
+
|
|
159
|
+
- `search_works` and `search_authors` answered a page sitting past the last row
|
|
160
|
+
by saying nothing in the catalogue carries those words. Asking for page 2 of
|
|
161
|
+
"Ruy Blas" reported an absence while twenty rows sat on page 1. A page beyond
|
|
162
|
+
the end and an absence are different answers, and only one of them means the
|
|
163
|
+
words match nothing. An empty page after the first is now read back at the
|
|
164
|
+
start of the same search: rows there mean the page sits past the end, and the
|
|
165
|
+
answer says so and points back to page 1. No row there means a real absence,
|
|
166
|
+
which is still stated plainly. A read-back that does not answer settles
|
|
167
|
+
neither, and the answer names both readings rather than choosing one.
|
|
168
|
+
- Page 1 is its own evidence and is never read twice, so a search that matches
|
|
169
|
+
nothing still costs one request.
|
|
170
|
+
- This states no total. The server does not ask the endpoint how many records
|
|
171
|
+
match, because a count on a search that does not rank would read as a measure
|
|
172
|
+
of the answer. Where the rows stop is a boundary it can point at.
|
|
173
|
+
|
|
5
174
|
## 1.0.0
|
|
6
175
|
|
|
7
176
|
First release.
|
package/LICENSE
CHANGED
|
File without changes
|
package/README.md
CHANGED
|
@@ -2,13 +2,16 @@
|
|
|
2
2
|
|
|
3
3
|
An MCP server for [data.bnf.fr](https://data.bnf.fr), the open catalogue of the
|
|
4
4
|
Bibliothèque nationale de France. Look up an author, find a work, list the
|
|
5
|
-
editions the BnF holds of it, and gather the links to
|
|
5
|
+
editions the BnF holds of it, list what a person wrote, and gather the links to
|
|
6
|
+
what has been digitised.
|
|
6
7
|
|
|
7
8
|
No API key. No account. Read-only.
|
|
8
9
|
|
|
9
10
|
[](https://www.npmjs.com/package/mcp-databnf)
|
|
10
11
|
[](https://github.com/smeet666/mcp-databnf/actions/workflows/ci.yml)
|
|
11
12
|
[](LICENSE)
|
|
13
|
+
[](https://registry.modelcontextprotocol.io/v0/servers?search=io.github.smeet666/mcp-databnf)
|
|
14
|
+
[](https://glama.ai/mcp/servers/smeet666/mcp-databnf)
|
|
12
15
|
|
|
13
16
|
---
|
|
14
17
|
|
|
@@ -20,7 +23,7 @@ web search does not: which of two people bearing one name wrote a given book,
|
|
|
20
23
|
what the BnF recorded as somebody's date and place of death, which editions of a
|
|
21
24
|
work exist and who printed them, and which of those have been digitised.
|
|
22
25
|
|
|
23
|
-
This server asks those questions for you, in
|
|
26
|
+
This server asks those questions for you, in seven tools, and reports what the
|
|
24
27
|
catalogue answers without adding to it.
|
|
25
28
|
|
|
26
29
|
## Install
|
|
@@ -42,7 +45,7 @@ npx mcp-databnf
|
|
|
42
45
|
}
|
|
43
46
|
```
|
|
44
47
|
|
|
45
|
-
## The
|
|
48
|
+
## The seven tools
|
|
46
49
|
|
|
47
50
|
| Tool | Answers |
|
|
48
51
|
| ---------------- | ------------------------------------------------------------------------------------------------------------------------- |
|
|
@@ -50,11 +53,12 @@ npx mcp-databnf
|
|
|
50
53
|
| `get_author` | Dates, places, occupation, language, country, Dewey class, and the same person in VIAF, IdRef, DBpedia, Wikidata and ISNI |
|
|
51
54
|
| `search_works` | Which works have these words in their title |
|
|
52
55
|
| `get_work` | Title, creators, date, language, form, subject, and whether the record is established or provisional |
|
|
56
|
+
| `list_works` | Which works the catalogue names a person the creator of, with the date and the form code it gives each |
|
|
53
57
|
| `list_editions` | Publisher, place, year, edition statement, extent, ISBN, catalogue link, and the Gallica link when there is one |
|
|
54
58
|
| `find_digitised` | Every digitised document the catalogue attaches to a person or a work, as links |
|
|
55
59
|
|
|
56
60
|
A typical exchange asks `search_authors` for a name, reads the rows, and passes
|
|
57
|
-
one identifier to `get_author` or `
|
|
61
|
+
one identifier to `get_author`, to `list_works`, or on to `list_editions`.
|
|
58
62
|
|
|
59
63
|
## What it does not do, and why
|
|
60
64
|
|
|
@@ -83,13 +87,40 @@ match. This server says so rather than inventing an order the catalogue does not
|
|
|
83
87
|
support, and it reports no total, because a total on a search that does not rank
|
|
84
88
|
reads as a measure of relevance.
|
|
85
89
|
|
|
90
|
+
**A name is matched letter for letter, and only among the people.** The index
|
|
91
|
+
compares the characters given against the characters a cataloguer entered, so
|
|
92
|
+
`Dostoïevski`, `Dostoievski` and `Dostoevskij` are three searches reaching three
|
|
93
|
+
different sets of records, and a handful of rows under one spelling is no
|
|
94
|
+
evidence about the others. `search_authors` says so on every answer, and it reads
|
|
95
|
+
the person records alone: an organisation or a conference is outside what it
|
|
96
|
+
looked at, so an answer holding no row is never a statement that the authority
|
|
97
|
+
file holds no such heading.
|
|
98
|
+
|
|
99
|
+
**A list of works is one link, not a bibliography.** `list_works` walks
|
|
100
|
+
`dcterms:creator`, the statement that ties a work to the person who made it.
|
|
101
|
+
That statement is real and it is partial. The catalogue credits a person on a
|
|
102
|
+
record in other ways, and the BnF holds printed editions whose work it has never
|
|
103
|
+
established as a record of its own, so a title missing from the list is not a
|
|
104
|
+
title the person did not write. Every answer says so, and none of them reports a
|
|
105
|
+
total: a count of what one link reaches would be read as a count of what somebody
|
|
106
|
+
wrote.
|
|
107
|
+
|
|
108
|
+
**A form code says what a work declares, and an empty one says nothing.** A work
|
|
109
|
+
record points at a term of the BnF's work-form vocabulary, and this dataset
|
|
110
|
+
publishes no label for those terms. So `roman`, `poesi` and `te` arrive as the
|
|
111
|
+
codes they are: some read as words and some do not. The catalogue also states no
|
|
112
|
+
form at all on a great many works, and that silence is a form unstated rather
|
|
113
|
+
than a genre denied. Keeping the rows that carry one code therefore finds the
|
|
114
|
+
works declaring it and never all the works of that form, which is why the codes
|
|
115
|
+
are reported and no filter is offered on them.
|
|
116
|
+
|
|
86
117
|
**It does not write biographies.** The field the BnF calls biographical
|
|
87
118
|
information is an occupation on most records: Rimbaud's says _Poète_, and that is
|
|
88
119
|
the whole of it. `get_author` returns that word and says what it is.
|
|
89
120
|
|
|
90
121
|
**It exposes no raw SPARQL tool.** An arbitrary query is an unbounded load on a
|
|
91
122
|
service a public institution pays for, and nothing here would control what the
|
|
92
|
-
caller wrote. Every query this server sends is one of
|
|
123
|
+
caller wrote. Every query this server sends is one of ten written in advance.
|
|
93
124
|
|
|
94
125
|
## The licence, and what it asks of you
|
|
95
126
|
|
|
@@ -205,8 +236,8 @@ retrieval.
|
|
|
205
236
|
|
|
206
237
|
Un serveur MCP pour [data.bnf.fr](https://data.bnf.fr), le catalogue ouvert de la
|
|
207
238
|
Bibliothèque nationale de France. Chercher un auteur, trouver une œuvre, lister
|
|
208
|
-
les éditions que la BnF en conserve,
|
|
209
|
-
numérisé.
|
|
239
|
+
les éditions que la BnF en conserve, lister ce qu'une personne a écrit, et
|
|
240
|
+
rassembler les liens vers ce qui a été numérisé.
|
|
210
241
|
|
|
211
242
|
Sans clé d'API. Sans compte. En lecture seule.
|
|
212
243
|
|
|
@@ -219,7 +250,7 @@ tel livre, ce que la BnF a enregistré comme date et lieu de mort de quelqu'un,
|
|
|
219
250
|
quelles éditions d'une œuvre existent et qui les a imprimées, et lesquelles ont
|
|
220
251
|
été numérisées.
|
|
221
252
|
|
|
222
|
-
Ce serveur pose ces questions pour vous, en
|
|
253
|
+
Ce serveur pose ces questions pour vous, en sept outils, et rapporte ce que le
|
|
223
254
|
catalogue répond sans y ajouter.
|
|
224
255
|
|
|
225
256
|
## Installation
|
|
@@ -239,16 +270,17 @@ npx mcp-databnf
|
|
|
239
270
|
}
|
|
240
271
|
```
|
|
241
272
|
|
|
242
|
-
## Les
|
|
273
|
+
## Les sept outils
|
|
243
274
|
|
|
244
|
-
| Outil | Répond à
|
|
245
|
-
| ---------------- |
|
|
246
|
-
| `search_authors` | Qui la BnF enregistre sous ce nom, et quelle notice est laquelle
|
|
247
|
-
| `get_author` | Dates, lieux, profession, langue, pays, indice Dewey, et la même personne dans VIAF, IdRef, DBpedia, Wikidata et ISNI
|
|
248
|
-
| `search_works` | Quelles œuvres portent ces mots dans leur titre
|
|
249
|
-
| `get_work` | Titre, auteurs, date, langue, forme, sujet, et si la notice est établie ou provisoire
|
|
250
|
-
| `
|
|
251
|
-
| `
|
|
275
|
+
| Outil | Répond à |
|
|
276
|
+
| ---------------- | ---------------------------------------------------------------------------------------------------------------------- |
|
|
277
|
+
| `search_authors` | Qui la BnF enregistre sous ce nom, et quelle notice est laquelle |
|
|
278
|
+
| `get_author` | Dates, lieux, profession, langue, pays, indice Dewey, et la même personne dans VIAF, IdRef, DBpedia, Wikidata et ISNI |
|
|
279
|
+
| `search_works` | Quelles œuvres portent ces mots dans leur titre |
|
|
280
|
+
| `get_work` | Titre, auteurs, date, langue, forme, sujet, et si la notice est établie ou provisoire |
|
|
281
|
+
| `list_works` | Quelles œuvres le catalogue attribue à une personne comme créatrice, avec la date et le code de forme qu'il leur donne |
|
|
282
|
+
| `list_editions` | Éditeur, lieu, année, mention d'édition, pagination, ISBN, lien catalogue, et le lien Gallica quand il existe |
|
|
283
|
+
| `find_digitised` | Tous les documents numérisés que le catalogue rattache à une personne ou à une œuvre, sous forme de liens |
|
|
252
284
|
|
|
253
285
|
## Ce qu'il ne fait pas, et pourquoi
|
|
254
286
|
|
|
@@ -278,6 +310,34 @@ Ce serveur le dit, plutôt que d'inventer un ordre que le catalogue ne porte pas
|
|
|
278
310
|
et il ne rapporte aucun total : sur une recherche qui ne classe pas, un total se
|
|
279
311
|
lit comme une mesure de pertinence.
|
|
280
312
|
|
|
313
|
+
**Un nom est apparié lettre à lettre, et seulement parmi les personnes.**
|
|
314
|
+
L'index compare les caractères donnés à ceux qu'un catalogueur a saisis :
|
|
315
|
+
`Dostoïevski`, `Dostoievski` et `Dostoevskij` sont donc trois recherches qui
|
|
316
|
+
atteignent trois ensembles de notices différents, et quelques lignes obtenues
|
|
317
|
+
sous une graphie ne disent rien des autres. `search_authors` l'énonce sur chaque
|
|
318
|
+
réponse, et il ne lit que les notices de personnes : un organisme ou un congrès
|
|
319
|
+
est hors de ce qu'il a regardé, si bien qu'une réponse sans ligne n'affirme
|
|
320
|
+
jamais que le fichier d'autorité ne tient pas une telle vedette.
|
|
321
|
+
|
|
322
|
+
**Une liste d'œuvres est un lien, pas une bibliographie.** `list_works` suit
|
|
323
|
+
`dcterms:creator`, l'énoncé qui rattache une œuvre à qui l'a faite. Cet énoncé
|
|
324
|
+
est réel et il est partiel : le catalogue crédite une personne sur une notice par
|
|
325
|
+
d'autres voies, et la BnF conserve des éditions imprimées dont l'œuvre n'a jamais
|
|
326
|
+
été établie comme notice à part entière. Un titre absent de la liste n'est donc
|
|
327
|
+
pas un titre que la personne n'a pas écrit. Chaque réponse le dit, et aucune ne
|
|
328
|
+
rapporte de total : un compte de ce qu'un seul lien atteint se lirait comme un
|
|
329
|
+
compte de ce que quelqu'un a écrit.
|
|
330
|
+
|
|
331
|
+
**Un code de forme dit ce qu'une œuvre déclare ; son absence ne dit rien.** Une
|
|
332
|
+
notice d'œuvre pointe vers un terme du vocabulaire des formes d'œuvre de la BnF,
|
|
333
|
+
et ce jeu de données n'en publie aucun libellé. `roman`, `poesi` et `te`
|
|
334
|
+
arrivent donc tels quels : certains se lisent comme des mots, d'autres non. Le
|
|
335
|
+
catalogue n'énonce par ailleurs aucune forme sur quantité d'œuvres, et ce
|
|
336
|
+
silence est une forme non déclarée, pas un genre exclu. Garder les lignes qui
|
|
337
|
+
portent un code trouve donc les œuvres qui le déclarent, jamais toutes les
|
|
338
|
+
œuvres de cette forme : les codes sont rapportés, et aucun filtre n'est offert
|
|
339
|
+
dessus.
|
|
340
|
+
|
|
281
341
|
**Il n'écrit pas de biographies.** Le champ que la BnF appelle information
|
|
282
342
|
biographique contient une profession sur la plupart des notices : celle de
|
|
283
343
|
Rimbaud dit _Poète_, et c'est tout. `get_author` rend ce mot et dit ce que c'est.
|
|
@@ -285,7 +345,7 @@ Rimbaud dit _Poète_, et c'est tout. `get_author` rend ce mot et dit ce que c'es
|
|
|
285
345
|
**Il n'expose aucun outil SPARQL brut.** Une requête arbitraire est une charge
|
|
286
346
|
non bornée sur un service qu'une institution publique paie, et rien ici ne
|
|
287
347
|
contrôlerait ce que l'appelant a écrit. Chacune des requêtes envoyées est l'une
|
|
288
|
-
des
|
|
348
|
+
des dix écrites à l'avance.
|
|
289
349
|
|
|
290
350
|
## La licence, et ce qu'elle vous demande
|
|
291
351
|
|
package/dist/bnf/client.d.ts
CHANGED
|
@@ -36,6 +36,14 @@ interface DigitisedLink {
|
|
|
36
36
|
ark: string;
|
|
37
37
|
/** The address to open, exactly as data.bnf.fr publishes it. */
|
|
38
38
|
url: string;
|
|
39
|
+
/**
|
|
40
|
+
* The view the address asks Gallica for, such as `thumbnail` or `f3.item`.
|
|
41
|
+
*
|
|
42
|
+
* Null when the address names the document itself. An address asking for a
|
|
43
|
+
* view opens that view, so the ARK beside it names the document a rendering
|
|
44
|
+
* was taken from rather than what a reader will see.
|
|
45
|
+
*/
|
|
46
|
+
rendering: string | null;
|
|
39
47
|
/**
|
|
40
48
|
* Which statement attached this link to the record.
|
|
41
49
|
*
|
|
@@ -130,6 +138,24 @@ interface WorkSummary {
|
|
|
130
138
|
status: "established" | "provisional";
|
|
131
139
|
sourceUrl: string;
|
|
132
140
|
}
|
|
141
|
+
/** One row of the works a person is named the creator of. */
|
|
142
|
+
interface AuthoredWork {
|
|
143
|
+
id: string;
|
|
144
|
+
title: string | null;
|
|
145
|
+
/** The date the record gives the work, as published. */
|
|
146
|
+
date: string | null;
|
|
147
|
+
/** The same date as a number, when the record states one. */
|
|
148
|
+
year: number | null;
|
|
149
|
+
/**
|
|
150
|
+
* Terms of the BnF work-form vocabulary the record points at, as the terms
|
|
151
|
+
* themselves. The vocabulary publishes no label for them in this dataset, so
|
|
152
|
+
* nothing here translates one: some read as words and some do not. An empty
|
|
153
|
+
* list is a form the record does not state.
|
|
154
|
+
*/
|
|
155
|
+
forms: string[];
|
|
156
|
+
status: "established" | "provisional";
|
|
157
|
+
sourceUrl: string;
|
|
158
|
+
}
|
|
133
159
|
/** One work, read in full. */
|
|
134
160
|
interface WorkDetail {
|
|
135
161
|
id: string;
|
|
@@ -203,8 +229,32 @@ interface Page<T> {
|
|
|
203
229
|
hasMore: boolean;
|
|
204
230
|
/** Rows the endpoint sent that could not be read, present only when some were. */
|
|
205
231
|
skipped?: number;
|
|
232
|
+
/**
|
|
233
|
+
* Whether the fixed window a text search reads off the index came back full.
|
|
234
|
+
*
|
|
235
|
+
* A full window bounds the answer: records carrying the words sit past it and
|
|
236
|
+
* were never read, so `hasMore` false means the rows of the window stop here
|
|
237
|
+
* rather than the catalogue holding nothing further. A window with room to
|
|
238
|
+
* spare holds everything that matched, which makes the answer both complete
|
|
239
|
+
* and repeatable.
|
|
240
|
+
*
|
|
241
|
+
* Null when the endpoint sent no occupancy, so nothing is claimed either way.
|
|
242
|
+
* Absent on a listing that reads no index window.
|
|
243
|
+
*/
|
|
244
|
+
indexWindowFull?: boolean | null;
|
|
206
245
|
}
|
|
207
246
|
|
|
247
|
+
/** How a piece of caller text became the terms the index was given. */
|
|
248
|
+
interface SearchReading {
|
|
249
|
+
/** The words the text was cut into, each one quoted into the query. */
|
|
250
|
+
words: string[];
|
|
251
|
+
/** The terms the index requires, after it splits inside a word. */
|
|
252
|
+
terms: string[];
|
|
253
|
+
/** The characters that marked nothing and were removed before the cutting. */
|
|
254
|
+
invisible: string[];
|
|
255
|
+
/** The visible characters that reached no term. */
|
|
256
|
+
setAside: string[];
|
|
257
|
+
}
|
|
208
258
|
type EntityKind = "ark" | "temp-work";
|
|
209
259
|
interface EntityId {
|
|
210
260
|
kind: EntityKind;
|
|
@@ -281,6 +331,21 @@ declare class BnfClient {
|
|
|
281
331
|
identify(input: string): EntityId;
|
|
282
332
|
/** The words a search will actually look for, which a tool reports back. */
|
|
283
333
|
words(input: string): string[];
|
|
334
|
+
/**
|
|
335
|
+
* The terms the index requires, which is what a row's presence rests on.
|
|
336
|
+
*
|
|
337
|
+
* The index splits inside a word at an apostrophe and at a hyphen, so a word
|
|
338
|
+
* a caller wrote as one can reach it as two, each required on its own.
|
|
339
|
+
*/
|
|
340
|
+
indexTerms(words: readonly string[]): string[];
|
|
341
|
+
/**
|
|
342
|
+
* How a piece of caller text became the terms the search was run with.
|
|
343
|
+
*
|
|
344
|
+
* A tool reports the reading back, so the same reading has to be the one the
|
|
345
|
+
* query was built from: the words, the terms the index makes of them, and
|
|
346
|
+
* the characters that reached neither.
|
|
347
|
+
*/
|
|
348
|
+
searchReading(input: string): SearchReading;
|
|
284
349
|
searchAuthors(name: string, limit: number, offset: number): Promise<Read<Page<AuthorSummary>>>;
|
|
285
350
|
getAuthor(id: EntityId): Promise<Read<AuthorDetail>>;
|
|
286
351
|
/**
|
|
@@ -293,6 +358,14 @@ declare class BnfClient {
|
|
|
293
358
|
private refuse;
|
|
294
359
|
searchWorks(title: string, limit: number, offset: number): Promise<Read<Page<WorkSummary>>>;
|
|
295
360
|
getWork(id: EntityId): Promise<Read<WorkDetail>>;
|
|
361
|
+
/**
|
|
362
|
+
* The works one person is named the creator of.
|
|
363
|
+
*
|
|
364
|
+
* The record is not checked here. A person the catalogue credits with nothing
|
|
365
|
+
* and an address that names no person both answer with no rows, and telling
|
|
366
|
+
* them apart costs a second query that only an empty answer needs.
|
|
367
|
+
*/
|
|
368
|
+
worksByAuthor(id: EntityId, limit: number, offset: number): Promise<Read<Page<AuthoredWork>>>;
|
|
296
369
|
listEditions(id: EntityId, limit: number, offset: number): Promise<Read<Page<Edition>>>;
|
|
297
370
|
/** Digitised documents attached to a work and to its editions. */
|
|
298
371
|
digitisedForWork(id: EntityId, limit: number): Promise<Read<Page<DigitisedLink>>>;
|