Escaping of escape symbols in the MetaData

[lenhard]

Fixes #2426 and is ready for review.

Old discussion

Attempts to fix #2426.

Is not yet functional, since the issue goes somewhat deeper and affects LaTeX-encoding things in all MetaData, also groups. The problem is line 64 in MetaDataSerializer:

 stringBuilder.append(StringUtil.quote(dataItem, MetaData.SEPARATOR_STRING, MetaData.ESCAPE_CHARACTER)).append(MetaData.SEPARATOR_STRING);

For every MetaData item, all special characters are quoted, including the quoting symbol (\). Thus, if there is a \ in the MetaData, it is quoted during serialization, resulting in \\, which is read as such and requoted on the next serialization, resulting in \\\\ and so on.

Unfortunately, unquoting the text during parsing and quoting it again during writing does not help, since a single \ will simply be lost. E.g.: unquoting \"U would result in "U, which upon quoting anew would stay "U.

The same problem should occur for groups, if I am not mistaken.

I now have the following options to resolve the problem:

1. Add special handling for content selectors to avoid quoting for them
2. Change StringUtil.quote to avoid quoting the quoting char (\)
3. Add another method in StringUtil that avoids quoting the quoting char and use it for MetaData in general.

I am afraid that changing StringUtil.quote has wider implications that I am not aware of., especially relating to groups Hence, there is a need for discussion.

• Change in CHANGELOG.md described
• Tests created for changes
• Manually tested changed features in running JabRef

[tobiasdiez]

I don't understand why unquoting on parse and quoting on writing does not work.
The input \"U will be written as \\\"U (as in the java string enquoting) and upon parsing be restored as \"U. Of course, a single backslash in the bib file gets lost but JabRef never should write just one backslash but always two.

As for the StringUtil.quote method, I think, you can edit it directly. If I remember correctly, then its purpose is to ensure that strings are properly encoded for writing them in the bib-file (this applies to all kind of metadata - groups are nothing special). Probably, it would be good to move this method actually to the metadata serializer / parser.

[lenhard]

@tobiasdiez You are correct, of course. If we avoid that JabRef writes a single backslash into the file, then everything works fine. My bad, for not getting this immediately. The current PR should be ready for review.

[tobiasdiez]

Build is fine expect a fetcher problem -> Merge