Unofficial support for Isabelle's Read-Eval-Print-Loop.

Features

• Python Client, easy for Machine Learning.
• Full support for all Isabelle commands
• Tracing the proof state and the output panel for each step of evaluation.
  • Plugin: collecting data from Isabelle's internal representations by plugins written in Isabelle/ML (example).
• Socket based remote communication
• Socket based Concurrency
• State rollback (example)
• Parsing terms & Retrival of lemmas (example)
• Sledgehammer (example)
• Retrivial of contextual information including assumptions, local facts, fixed terms, fixed types, and goals, in either the pretty-printed form or S expression (example).

We only support Isabelle2023 and Isabelle2024.

Ensuring <ISABELLE-BASE-DIRECTORY>/bin is in your $PATH environment

git clone https://github.com/xqyww123/Isa-REPL.git
cd Isa-REPL
git checkout $(isabelle version) # Error can raise if you are using an unsupported version of Isabelle
isabelle components -u .
pip install IsaREPL

# merely an example:
./repl_server.sh 127.0.0.1:6666 HOL ./tmp

Run ./repl_server.sh to see the full explanation of the arguments and options.

Every interface of our client is will documented (it is highly recommended to read our source). Some examples are given in the example folder.

# Run an example
./examples/example_eval.py 127.0.0.1:6666

To evaluate a whole theory file

./examples/eval_file.py 127.0.0.1:6666 $(isabelle getenv -b ISABELLE_HOME)/src/HOL/List.thy

All documents are given as docstrings attaching to Python client interfaces, see the source.

This REPL supports concurrent evaluation of multiple files. However, to ensure proper error tracking and result tracing, it disables Isabelle's concurrency between commands within a single theory file. While Isabelle typically executes by commands asynchronously (continuing to the next command before completion), this REPL enforces synchronous evaluation. Each command must complete before the next begins, allowing for accurate capture and reporting of any failures.

Sledgehammer (via our auto_sledgehamemr wrapper, see example) is still concurrent.

It is possible to implement a client in other languages. However, the document for the communication protocol is not provided and I refer you to read our source code, as it should be simple enough.

Feel free to open any GitHub issue if you have any feature requests.

I am not a professional python developer, so contribution is highly welcome to enrich the client's features.