Refuse to create logfile when the name doesn't end in .spl

[nresare]

Instead of just complaining :)

[nailor]

Just out of curiosity: does the implementation really rely somewhere, in some client lib, that the file ends up in the .spl?

[nresare]

I don't think so. It's more a convenience for new users that the writehash CLI tool expects an .spl. API users are on their own and can call the files whatever they like.

[nailor]

I still kind of liked the warning mode, instead of completely blocking you to write a file if the extension is wrong

[spkrka]

It's somewhat mixed:

• The C and Python API:s don't assume any specific file endings. You can use what you want.
• The get operation on commandline assumes .spi and .spl file endings, mostly so you don't have to input both files.
• The writehash operation on commandline assumes .spi and .spl in order to be able to know what the call the newly created index file.
• The java API mostly enforces the .spi and .spl file endings.

I am not sure what approach is best and if we should enforce it everywhere.

[nailor]

But if this is already the behavior in command line, then forcing this makes sense IMO.

What to do with the APIs, I don't know

[rohansingh]

I'm a big fan of being able to name files whatever you want.

[spkrka]

I think I may agree. I think I'll leave the commandline tool to be as permissive as already is, and possibly also extend the Java API to be able to specify explicit filenames instead of following the convention (as soon as someone actually decides they have that need).

The commands that already assume the convention only do so to simplify the usage, if someone strictly wants some other filenames they can write a minimal commandline tool using the python (or c) api.

[nresare]

I'm fine with a warning as well (or the current behavior)

[nresare]

If we decide to go with freedom to the user (sounds nice, doesn't it?) I would propose that we changed writehash to not require .spl.

[spkrka]

Yes, but to do that, the commandline tool needs to take one additional argument to know both the log file and the index file.

Maybe we could add an optional parameter to specify index file, and if that's specified we drop the requirement that the log file ends with .spl

[Qix-]

I'm also in the camp of naming the files whatever you want. It's either enforce it library-wide, though, or don't enforce it at all. As a new Sparkey user, I ran into this as well - I could create a new logfile with whatever name I wanted, but then when I went to write to it it errored out. Was confused for a moment and it seemed less elegant than it probably should.

[spkrka]

This PR is getting a bit stale so we can't merge it regardless, until it's been updated to latest master.

I am not opposed to letting you choose filenames, but we'll need to fix that in all places so it becomes consistent. For backwards compatibility this means we should add more options to the command line for the affected

For sparkey get we need to add sparkey get [-l logfile] indexfile <key> instead of assuming that the log file ends with .spl.

For sparkey writehash we need to add sparkey writehash [-i indexfile] logfile instead of assuming that the index file ends with .spi.

For sparkey rewrite we need to add support for sparkey rewrite --input-index index-file --input-log log-file --output-index index-file --output-log log-file in addition to the existing syntax.

I think this problem only affects the commandline utility, the internals should already be explicit about these things.

I don't have much time at the moment to do this, but if someone else does I am happy to review it.