zotero/test/tests/fulltextTest.js
Dan Stillman 10181f7f56 Fix sync error on missing full-text
If a version is returned for an item's full-text content but a 404 is returned
for the content itself (because it's missing in Elasticsearch for some reason),
don't throw an error.

Also remove legacy array comprehensions in fulltext and syncFullTextEngine test
files, which apparently weren't being run.
2016-04-30 01:24:33 -04:00

213 lines
6.4 KiB
JavaScript

describe("Zotero.Fulltext", function () {
var win, pdfToolsVersion;
before(function* () {
// Hidden browser, which requires a browser window, needed for charset detection
// (until we figure out a better way)
win = yield loadBrowserWindow();
pdfToolsVersion = Zotero.isWin ? '3.02a' : '3.04';
});
after(function () {
if (win) {
win.close();
}
});
describe("#indexItems()", function () {
before(function* () {
yield Zotero.Fulltext.downloadPDFTool('info', pdfToolsVersion);
yield Zotero.Fulltext.downloadPDFTool('converter', pdfToolsVersion);
});
beforeEach(function () {
Zotero.Prefs.clear('fulltext.textMaxLength');
Zotero.Prefs.clear('fulltext.pdfMaxPages');
});
after(function () {
Zotero.Prefs.clear('fulltext.textMaxLength');
Zotero.Prefs.clear('fulltext.pdfMaxPages');
});
it("should index a text file by default", function* () {
var item = yield importFileAttachment('test.txt');
assert.equal(
(yield Zotero.Fulltext.getIndexedState(item)),
Zotero.Fulltext.INDEX_STATE_INDEXED
);
})
it("should skip indexing of a text file if fulltext.textMaxLength is 0", function* () {
Zotero.Prefs.set('fulltext.textMaxLength', 0);
var item = yield importFileAttachment('test.txt');
assert.equal(
(yield Zotero.Fulltext.getIndexedState(item)),
Zotero.Fulltext.INDEX_STATE_UNINDEXED
);
})
it("should index a PDF by default", function* () {
var item = yield importFileAttachment('test.pdf');
assert.equal(
(yield Zotero.Fulltext.getIndexedState(item)),
Zotero.Fulltext.INDEX_STATE_INDEXED
);
})
it("should skip indexing of a PDF if fulltext.textMaxLength is 0", function* () {
Zotero.Prefs.set('fulltext.textMaxLength', 0);
var item = yield importFileAttachment('test.pdf');
assert.equal(
(yield Zotero.Fulltext.getIndexedState(item)),
Zotero.Fulltext.INDEX_STATE_UNINDEXED
);
})
it("should skip indexing of a PDF if fulltext.pdfMaxPages is 0", function* () {
Zotero.Prefs.set('fulltext.pdfMaxPages', 0);
var item = yield importFileAttachment('test.pdf');
assert.equal(
(yield Zotero.Fulltext.getIndexedState(item)),
Zotero.Fulltext.INDEX_STATE_UNINDEXED
);
})
})
describe("#downloadPDFTool()", function () {
it("should install the PDF tools", function* () {
yield Zotero.Fulltext.uninstallPDFTools();
assert.isFalse(Zotero.Fulltext.pdfInfoIsRegistered());
var dataDir = Zotero.getZoteroDirectory().path;
var execFileName = Zotero.Fulltext.pdfInfoFileName;
var execPath = OS.Path.join(dataDir, execFileName);
var versionFileName = execFileName + '.version';
var versionPath = OS.Path.join(dataDir, versionFileName);
var scriptExt = Zotero.isWin ? 'vbs' : 'sh';
var scriptPath = OS.Path.join(dataDir, 'pdfinfo.' + scriptExt);
var scriptContents = yield Zotero.File.getContentsFromURLAsync(
'resource://zotero/redirect.' + scriptExt
);
var cacheExecPath = OS.Path.join(
getTestDataDirectory().path, "pdf", pdfToolsVersion, execFileName
);
// Delete existing files
try {
yield OS.File.remove(execPath);
}
catch (e) {}
try {
yield OS.File.remove(versionPath);
}
catch (e) {}
try {
yield OS.File.remove(scriptPath);
}
catch (e) {}
yield Zotero.Fulltext.downloadPDFTool('info', pdfToolsVersion);
assert.ok(Zotero.Fulltext.pdfInfoIsRegistered());
assert.equal(
(yield Zotero.File.getBinaryContentsAsync(cacheExecPath)),
(yield Zotero.File.getBinaryContentsAsync(execPath))
);
if (!Zotero.isWin) {
assert.equal((yield OS.File.stat(execPath)).unixMode, 0o755);
}
assert.equal(
(yield Zotero.File.getContentsAsync(versionPath)),
pdfToolsVersion
);
//Temp: disabled on Windows
if (!Zotero.isWin) {
assert.equal(
(yield Zotero.File.getContentsAsync(scriptPath)),
scriptContents
);
assert.equal((yield OS.File.stat(scriptPath)).unixMode, 0o755);
}
yield uninstallPDFTools();
assert.isFalse(Zotero.Fulltext.pdfInfoIsRegistered());
})
})
describe("#getUnsyncedContent()", function () {
before(function* () {
yield installPDFTools();
})
after(function* () {
yield uninstallPDFTools();
})
it("should get content that hasn't been uploaded", function* () {
var toSync = [];
var group = yield getGroup();
var add = Zotero.Promise.coroutine(function* (options = {}) {
let item = yield createDataObject('item', { libraryID: options.libraryID });
let attachment = new Zotero.Item('attachment');
if (options.libraryID) {
attachment.libraryID = options.libraryID;
}
attachment.parentItemID = item.id;
attachment.attachmentLinkMode = 'imported_file';
attachment.attachmentContentType = 'text/plain';
attachment.attachmentCharset = 'utf-8';
attachment.attachmentFilename = 'test.txt';
if (options.synced) {
attachment.synced = true;
}
yield attachment.saveTx();
yield Zotero.Attachments.createDirectoryForItem(attachment);
let path = attachment.getFilePath();
let content = new Array(10).fill("").map(x => Zotero.Utilities.randomString()).join(" ");
yield Zotero.File.putContentsAsync(path, content);
if (!options.skip) {
toSync.push({
item: attachment,
content,
indexedChars: content.length,
indexedPages: 0
});
}
});
yield add({ synced: true });
yield add({ synced: true });
// Unsynced attachment shouldn't uploaded
yield add({ skip: true });
// Attachment in another library shouldn't be uploaded
yield add({ libraryID: group.libraryID, synced: true, skip: true });
// PDF attachment
var pdfAttachment = yield importFileAttachment('test.pdf');
pdfAttachment.synced = true;
yield pdfAttachment.saveTx();
toSync.push({
item: pdfAttachment,
content: "Zotero [zoh-TAIR-oh] is a free, easy-to-use tool to help you collect, "
+ "organize, cite, and share your research sources.\n\n",
indexedChars: 0,
indexedPages: 1
});
yield Zotero.Fulltext.indexItems(toSync.map(x => x.item.id));
var data = yield Zotero.FullText.getUnsyncedContent(Zotero.Libraries.userLibraryID);
assert.lengthOf(data, 3);
let contents = toSync.map(x => x.content);
for (let d of data) {
let pos = contents.indexOf(d.content);
assert.isAbove(pos, -1);
assert.equal(d.indexedChars, toSync[pos].indexedChars);
assert.equal(d.indexedPages, toSync[pos].indexedPages);
}
})
})
})