mirror of
https://github.com/overleaf/overleaf.git
synced 2024-11-21 20:47:08 -05:00
114 lines
2.7 KiB
JavaScript
114 lines
2.7 KiB
JavaScript
|
const Path = require('path')
|
||
|
|
||
|
/**
|
||
|
* Parse output from the `synctex view` command
|
||
|
*/
|
||
|
function parseViewOutput(output) {
|
||
|
return _parseOutput(output, (record, label, value) => {
|
||
|
switch (label) {
|
||
|
case 'Page':
|
||
|
_setIntProp(record, 'page', value)
|
||
|
break
|
||
|
case 'h':
|
||
|
_setFloatProp(record, 'h', value)
|
||
|
break
|
||
|
case 'v':
|
||
|
_setFloatProp(record, 'v', value)
|
||
|
break
|
||
|
case 'W':
|
||
|
_setFloatProp(record, 'width', value)
|
||
|
break
|
||
|
case 'H':
|
||
|
_setFloatProp(record, 'height', value)
|
||
|
break
|
||
|
}
|
||
|
})
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parse output from the `synctex edit` command
|
||
|
*/
|
||
|
function parseEditOutput(output, baseDir) {
|
||
|
return _parseOutput(output, (record, label, value) => {
|
||
|
switch (label) {
|
||
|
case 'Input':
|
||
|
if (Path.isAbsolute(value)) {
|
||
|
record.file = Path.relative(baseDir, value)
|
||
|
} else {
|
||
|
record.file = value
|
||
|
}
|
||
|
break
|
||
|
case 'Line':
|
||
|
_setIntProp(record, 'line', value)
|
||
|
break
|
||
|
case 'Column':
|
||
|
_setIntProp(record, 'column', value)
|
||
|
break
|
||
|
}
|
||
|
})
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Generic parser for synctex output
|
||
|
*
|
||
|
* Parses the output into records. Each line is split into a label and a value,
|
||
|
* which are then sent to `processLine` for further processing.
|
||
|
*/
|
||
|
function _parseOutput(output, processLine) {
|
||
|
const lines = output.split('\n')
|
||
|
let currentRecord = null
|
||
|
const records = []
|
||
|
for (const line of lines) {
|
||
|
const [label, value] = _splitLine(line)
|
||
|
|
||
|
// A line that starts with 'Output:' indicates a new record
|
||
|
if (label === 'Output') {
|
||
|
// Start new record
|
||
|
currentRecord = {}
|
||
|
records.push(currentRecord)
|
||
|
continue
|
||
|
}
|
||
|
|
||
|
// Ignore the line if we're not in a record yet
|
||
|
if (currentRecord == null) {
|
||
|
continue
|
||
|
}
|
||
|
|
||
|
// Process the line
|
||
|
processLine(currentRecord, label, value)
|
||
|
}
|
||
|
return records
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Split a line in label and value components.
|
||
|
*
|
||
|
* The components are separated by a colon. Note that this is slightly
|
||
|
* different from `line.split(':', 2)`. This version puts the entirety of the
|
||
|
* line after the colon in the value component, even if there are more colons
|
||
|
* on the line.
|
||
|
*/
|
||
|
function _splitLine(line) {
|
||
|
const splitIndex = line.indexOf(':')
|
||
|
if (splitIndex === -1) {
|
||
|
return ['', line]
|
||
|
}
|
||
|
return [line.slice(0, splitIndex).trim(), line.slice(splitIndex + 1).trim()]
|
||
|
}
|
||
|
|
||
|
function _setIntProp(record, prop, value) {
|
||
|
const intValue = parseInt(value, 10)
|
||
|
if (!isNaN(intValue)) {
|
||
|
record[prop] = intValue
|
||
|
}
|
||
|
}
|
||
|
|
||
|
function _setFloatProp(record, prop, value) {
|
||
|
const floatValue = parseFloat(value)
|
||
|
if (!isNaN(floatValue)) {
|
||
|
record[prop] = floatValue
|
||
|
}
|
||
|
}
|
||
|
|
||
|
module.exports = { parseViewOutput, parseEditOutput }
|