Citation key, make it possible for an empty field to fall back to a custom field

[lantyn]

Nowadays, most articles from online database already have their unique identifiers.
I personally prefer to use [authorsAlpha][eprint] as the default pattern for citation key.
It expands the default [author][year] pattern (the information of [year] is already contained in [eprint]); moreover, it automatically avoid duplicate keys.
However, it is not compatible with old articles with no [eprint] field. So I hope that when [eprint] is empty, it automatically falls back to [year]. Currently JabRef has the feature to replace an empty field with a custom string, which is actually not informative at all and thus not practical. Also when [author] is empty then [editor] is automatically used. So why not make this available to all empty fields with customizable fallback field? The syntax may be simply [eprint:[year]].

[Siedlerchr]

This should be possible at least to define a fallback
https://docs.jabref.org/setup/citationkeypatterns#citation-key-patterns

:(x): The string between the parentheses will be inserted if the field marker preceding this modifier resolves to an empty value. The placeholder x may be any string. For instance, the marker [VOLUME:(unknown)] will return the entry's volume if set, and the string unknown if the entry's VOLUME field is not set

@k3KAW8Pnf7mkmdSMPHz27 Do you know if it's possible to define a fallback field?

[k3KAW8Pnf7mkmdSMPHz27]

Nowadays, most articles from online database already have their unique identifiers.

You are referring to eprint?

Currently JabRef has the feature to replace an empty field with a custom string,

It would be better if it supported fields/bracketed expressions as well, yup.

You might be able to workaround the issue using the replace regular expression feature with an empty string. I am not familiar enough with regex and the year/eprint fields to come up with something good, but for the pattern [author]|[eprint][year] something along the lines of (?<=|)\d\d\d\d$ (replace the last 4 digits if they are not preceded by |) might work (even if it doesn't give the exact results that you want)?

So why not make this available to all empty fields with customizable fallback field?

I like the idea ``

The syntax may be simply [eprint:[year]]

I'd recommend [eprint:([year])], in that case, for consistency. I'd be happy to review/help out if you are interested in implementing this.

jabref/src/main/java/org/jabref/logic/citationkeypattern/BracketedPattern.java

Lines 549 to 553 in b0af521

	} else if (!modifier.isEmpty() && (modifier.length() >= 2) && (modifier.charAt(0) == '(') && modifier.endsWith(")")) {
	// Alternate text modifier in parentheses. Should be inserted if the label is empty
	if (label.isEmpty() && (modifier.length() > 2)) {
	resultingLabel = modifier.substring(1, modifier.length() - 1);
	}

Is the (current) replacement modifier.

Do you know if it's possible to define a fallback field?

As far as I know, no =/

[lantyn]

Nowadays, most articles from online database already have their unique identifiers.

Yes, eprint or even doi. I don't like the [year] field because new articles tend to have both a preprint and a reprint with different [year]. When I update the reprint info from a preprint, the year changes, which then messes up the bibtex key and the pdf file names. I then decided to use eprint instead of year which is unique and persistent.

I do have my personal workaround via regex.
I set the pattern [authorsAlpha]:[year]:[eprint], then replace :(.*):$|:.*:(\S) with $1$2.
So when [eprint] in not empty I get [authorsAlpha][eprint], otherwise I get [authorsAlpha][year].
Such hacking should work for customizing a fallback of any empty field.

However, I suppose that in the current version you can do regex replace only once. I would be already happy if it was allowed to add several regex replacement rules.

[k3KAW8Pnf7mkmdSMPHz27]

Such hacking should work for customizing a fallback of any empty field.

Nice! Might be that others can use this meanwhile.

However, I suppose that in the current version you can do regex replace only once. I would be already happy if it was allowed to add several regex replacement rules.

Agreed.

[SuXiChangZhen]

@k3KAW8Pnf7mkmdSMPHz27 Hi, I am interested in this issue. Where can I get started and What is the expected behavior of citation key generation? Thank you.

[k3KAW8Pnf7mkmdSMPHz27]

Hello @SuXiChangZhen, thank you for your interest!

Pretty much all approaches start with the "alternative text modifier" as described above,

jabref/src/main/java/org/jabref/logic/citationkeypattern/BracketedPattern.java

Lines 549 to 553 in b0af521

	} else if (!modifier.isEmpty() && (modifier.length() >= 2) && (modifier.charAt(0) == '(') && modifier.endsWith(")")) {
	// Alternate text modifier in parentheses. Should be inserted if the label is empty
	if (label.isEmpty() && (modifier.length() > 2)) {
	resultingLabel = modifier.substring(1, modifier.length() - 1);
	}

It is listed as the last modifier in the JabRef documentation.

There isn't necessarily one "right" answer to this issue. The code is old and has a few violations against good coding principles making it harder to deal with it. Mostly I'd advise against trying to refactor the whole BracketedPattern class because it would be very time-consuming. I'd guest the best effort/utility trade-off would be to extend the method injection code found among other places in,

jabref/src/main/java/org/jabref/logic/citationkeypattern/CitationKeyGenerator.java

Lines 195 to 220 in 807f85e

	return expandBrackets(citationKeyPattern.get(0), expandBracketContent(entry));
	}
	/**
	* A helper method to create a {@link Function} that takes a single bracketed expression, expands it, and cleans the key.
	*
	* @param entry the {@link BibEntry} that a citation key is generated for
	* @return a cleaned citation key for the given {@link BibEntry}
	*/
	private Function<String, String> expandBracketContent(BibEntry entry) {
	Character keywordDelimiter = citationKeyPatternPreferences.getKeywordDelimiter();
	return (String bracket) -> {
	String expandedPattern;
	List<String> fieldParts = parseFieldAndModifiers(bracket);
	expandedPattern = removeUnwantedCharacters(getFieldValue(entry, fieldParts.get(0), keywordDelimiter, database), unwantedCharacters);
	// check whether there is a modifier on the end such as
	// ":lower":
	if (fieldParts.size() > 1) {
	// apply modifiers:
	expandedPattern = applyModifiers(expandedPattern, fieldParts, 1);
	}
	return cleanKey(expandedPattern, unwantedCharacters);
	};
	}

so that

jabref/src/main/java/org/jabref/logic/citationkeypattern/BracketedPattern.java

Line 537 in 807f85e

static String applyModifiers(final String label, final List<String> parts, final int offset) {

can parse and expand whatever is between the parentheses, i.e., make the same type of call as in

jabref/src/main/java/org/jabref/logic/citationkeypattern/CitationKeyGenerator.java

Line 195 in 807f85e

return expandBrackets(citationKeyPattern.get(0), expandBracketContent(entry));

Does that sort of make sense?

You are also free to solve it in some completely different way.

expected behavior of citation key generation

I am not sure that I understand your question. The bracketed patterns were originally intended to create a citation key from bibentries by accessing its fields, (e.g., [TITLE] accesses StandardField.TITLE) with some special keywords creating shorter/easier-to-read citation keys (e.g. [title]). The issue here is regarding accessing a different field if [title] resolves into an empty string. At the moment it is only possible to use a string instead, e.g., [TITLE:(NoTitle)], to generate the citation key "NoTitle" if the bibentry has no title. But it would be good to be able to replace it with a different field instead (e.g., [EPRINT:([YEAR])], if there is no EPRINT field, try to use the YEAR field instead),

[SuXiChangZhen]

Thank you for your advice. I think I understand the need. I will make a pr soon.