How to deal with unsupported language projects

[cx-alberto-simoes]

I am writing a generic tool, and we need to have serena working with projects of not supported programming languages. For instance, we have a project only with SQL files.
Currently, serena fails initialization because there is no supported language.
Is there a way to force serena to accept a project even without any supported language?

[MischaPanch]

I am not sure what you mean. Serena will initialize with the languages that are explicitly configured in project.yml. Of course, manually adding an unsupported language there will lead to a crash, but that's something the user has to do by hand.

[MischaPanch]

Btw, if you use Serena through the JetBrains plugin, all such issues will disappear - Serena will have access to all symbols indexed by the JetBrains IDE

[cx-alberto-simoes]

@MischaPanch this is a standalone tool, so not inside jetbrains.

If there is NO language supported in the project (project only with sql files, for instance), how to create a proper project.yml that allows read_file to work on those files?

[MischaPanch]

In that case, 95% of Serena's functionality is disabled so it just doesn't make sense to use it at all. I suggest switching to desktop commander or something else then

[MischaPanch]

You can configure the markdown language as a workaround or something like that if you really want to start Serena

[cx-alberto-simoes]

@MischaPanch having two separate backends does not make sense. Also, who knows if in the future Serena doesn't add support for the missing languages?

[opcode81]

@cx-alberto-simoes if your project matches none of the supported languages, the workaround is to just use any supported language and specify that language explicitly at project creation. I would recommend python as it requires no additional dependencies or anything.

<serena> project create --language python /path/to/your/project

You will then be able to apply all of Serena's non-symbolic tools without any problems. The Python language server will spawn, but it will not actually be used for anything.

[MischaPanch]

That's what I suggested above, proposing the markdown language specifically

[cx-alberto-simoes]

Would something like this be acceptable by Serena team? #1017

[opcode81]

For a multi-language project, auto-detection will only use the dominant language; the use of multiple languages is always user-controlled. When using the explicit project creation process, the user is prompted for the additional languages to support.

Without the user choosing the languages manually, a workaround would be to add a dummy file to each project (e.g. dummy.md or dummy.py). This will achieve that the dominant language will still be correctly auto-detected if the project contains any supported source files and will always use a fallback language server if not (i.e. markdown or Python).

[opcode81]

Note that it was an explicit decision to activate only the main language by default. Many projects contain a large number of different languages, but most of them are not essential to core development. Every language server uses significant resources and this could become a problem if we just activated e.g. 5 of them without even asking. Also, many language servers require system-level installations, and the failure to start a language server is a critical error (as the assumption is that every language that is enabled must be supported).

This is a consistent way of thinking about things, which, however, assumes a user interacting and making decisions.
Your automated use case is different from the one Serena is designed for.

[cx-alberto-simoes]

Thanks for the explanation.
Activating only the dominant language is also an issue for us.
Will try to understand a workaround.

[cx-alberto-simoes]

Just an extra question, if I activate a language, I am able to do text-search or list files for other language, or they are restricted to the active languages?

[opcode81]

You can do everything except use the symbolic tools like find_symbol or replace_symbol_body: Those tools will only find/work for symbols in the languages you have activated.