allows deserealizing np.float64

[Harshghub]

No description provided.

[mosmuell]

Hey @Harshghub,
Thanks for this. For now, I have been avoiding adding support for numpy types entirely as this brings a bit of complexity with it, e.g.

• I would assume that serialised numpy floats could be derserialised into the same type again.
• other numpy types (int, bool, ...) should probably also be added
• we should make sure that numpy does not become a direct dependency of pydase. It should only be imported where necessary, in which case the user already installed numpy.
• As we have two different types of clients (python clients, web clients) we should think about how they handle numpy types. For example, if you expose a numpy float through a property, changing this in the frontend should not change it to a simple float. When you change it through the python client, however, the type could be overwritten... Also, the python client needs to have numpy installed when interacting with a server exposing a numpy type

That's just some thoughts on numpy types and why I didn't add support for them yet.

What are your thoughts on this?

[Harshghub]

Hello Mose, Thank you very much for your message. I understand that you have thought very deeply into this matter. Right now what I implemented is a workaround for the old measurement scripts which used to work on tiqi-plugins to work with pydase. I have discussed your points with Bahadir and he told me that you guys will be meeting soon and he can discuss these points with you. Maybe I will send you a follow-up after you guys have had the discussion. Right now, I also made changes to pydase_service_base in the ionizer class. Connecting with Ionizer was giving errors whenever the code tried to notify the ionizer for changes in DataService objects. The error was of the "Cannot serialize ___ type objects" kind. Currently this is also a workaround and maybe we can think of a more elegant and pythonian solution. Thank you very much for looking into this matter and giving time even after your contract. Best regards Harsh Vardhan Upadhyay

…

On Mon, Sep 29, 2025 at 5:43 PM Mose Müller ***@***.***> wrote: *mosmuell* left a comment (tiqi-group/pydase#259) <#259 (comment)> Hey @Harshghub <https://github.com/Harshghub>, Thanks for this. For now, I have been avoiding adding support for numpy types entirely as this brings a bit of complexity with it, e.g. - I would assume that serialised numpy floats could be derserialised into the same type again. - other numpy types (int, bool, ...) should probably also be added - we should make sure that numpy does not become a direct dependency of pydase. It should only be imported where necessary, in which case the user already installed numpy. - As we have two different types of clients (python clients, web clients) we should think about how they handle numpy types. For example, if you expose a numpy float through a property, changing this in the frontend should not change it to a simple float. When you change it through the python client, however, the type could be overwritten... Also, the python client needs to have numpy installed when interacting with a server exposing a numpy type That's just some thoughts on numpy types and why I didn't add support for them yet. What are your thoughts on this? — Reply to this email directly, view it on GitHub <#259 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AP6XVXXPLQFODEOK5LD4SED3VFHSZAVCNFSM6AAAAACHZLJY3SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNBXG4ZTCNBZGA> . You are receiving this because you were mentioned.Message ID: ***@***.***>