Nice. Hm, what to do when two entries share a line? E.g. @article{key1, title={a}}@article{key2, title={b}}?

In that case, the current implementation fails.

But that gives me an idea: Isn't @ a valid delimiter that is only used for setting bibtex entry types, but forbidden in other parts of the file? If so, I could use it for reading the file, instead of the newline character. This is very easy to do with a Scanner and would make things way easier.

There is the issue of comments, which potentially can contain @. But as far as I know, everything outside of @xxx{..} is a comment.

The @ symbol can also occur in an entry, for example in an email-address. The current implementation has also the problem that it fails if the entry has no bibtex key. I fear there is no easy way to find an entry in a bibtex file without actually parsing it completely.

Isn't it possible to add a reference to the source line and column during parsing? This should be offered by ANTLR somehow, isn't it?

I assume that JBibTeX. Doesn't support that. Refs #123.

@oscargus: that would not be a problem. It's more of a problem if an @ is within an entry

@koppor: This is work in progress. I'll start commenting when I have something closer to finalization. Do we want to parse the file twice? If so, then jbibtex might be an option. Can you tell me if jbibtex writes back an entry in exactly the same format (up to every line feed character) as it was read?

@tobiasdiez Thanks, I am aware of that. This is work in progress. I'd first like to arrive at a solution that does the round-trip for a well-structured database. Then, I'll modify it to match incomplete entries.

Ok, now this works nicely with my large local database that doesn't include analomies.

Now comes the hard part: Properly reading and writing files with strange content. This is also the reason why some tests are failing at the moment. I might need to discuss some of these aspects and will ping you in this thread if I am unsure.

Dammit, I wanted to get some work done today and now I am only coding JabRef...

Anyways, this seems to work (and passes the tests)! I ditched my previous parsing logic and rewrote it in integration with the BibtexParser, which wasn't as hard as I claimed above (my bad).

I am now able to store each entry exactly as it is in the file. Each entry serialization contains all text from its end up to the end of previous entry. The main problem with this is that with this logic, the first entry also stores the file header and then the header is serialized twice when rewriting the file. I was able to get around that with some hacking, but there might still be edge cases I am not yet aware of.

To sum up, this needs some more manual testing, but should be pretty close. Feedback is welcome :)

Tested it manually with some entries and it works smoothly!

Good job! 👍

Should we really introduce this in v3.0 already? I'm not entirely sure it works on all occaison. For instance, I haven't tested it with strings in the bib file so far.

e.g. https://github.com/JabRef/jabref/blob/master/src/main/java/net/sf/jabref/migrations/FileLinksUpgradeWarning.java#L65

@matthiasgeiger But that does not really depend on the version information. The version is checked, but also every file written by JabRef 3.0 is marked as problematic. The file links upgrade really depends on whether the file contains fields named pdf or ps. It wouldn't make a difference isActionNecessary always returned true, regardless of the version.

Anyways, if you really really want the version parsing, we can keep it.

okay... I have to confess that I only performed a "find usages" search without truly checking whether the usages are really needed.

So feel free to remove those lines ;-)

I removed the old meta flag and version parsing. Version headers are now simply removed on rewrite. File upgrade functionality ultimately depends on preferences (and on whether JabRef finds errors in the file to begin with).

So, from my point of view this PR is complete and ready to merge.

@koppor and I did some more testing with more advanced things like meta-data and it seems fine. I'm not guaranteeing the absence of bugs, but there should be no more obvious mistakes.

Somebody else can take a final look and merge this with master.

Looks good to me. It worked with my large .bib file without any issues.

@simonharrer Could you check what happens if you change the sort ordering of the file? Change the ordering. If everything is OK, change the ordering and change an entry. 🐇 I it doesn't work, maybe we drop the ordering possibilities of entries within JabRef now or leave that as open issue.

What do you think is the problem here? When I change the sort ordering, the order of the entries is changed in the file. But nothing else is. And if I also change an entry, this entry is changed as well. I do not get what you imply.