From Newsgroup: comp.lang.python


We present a Prolog transducer dubbed NetFish
that can be used to translate Java code into C#
code. NetFish can be built with a non-standard
version of DCG, extending the notion of semi-context
from terminals to non-terminals. NetFish can be
easily run over arbitrary long files
with little memory.

NetFish uses a sliding window along an input text.
It then applies the given compiled DSL rules in a
cascading style. We used various LLMs such as
ChatGPT and DeepSeek to advice us in the rules,
but such a process is currently not integrated.
The output can be run with .NET 9.0.

See also:

NetFish Transducer in Dogelog Player
https://medium.com/2989/9d392937c1e3

Mild Shock schrieb:

Dear All,

We are happy to announce a new edition
of the Dogelog Player:

- Enhanced GC:
To lift the native stack limitations, we
opted for a marking algorithm based on
Peter Deutschs algorithm E as found in
Donald Knuths "The Art of Computing Programming"
book. Our variant uses an int field that was
anyway recently introduced for Prolog
compound coloring, so that no extra space
was introduced in this release.

- Enhanced Binary-Ops:
By adopting the pointer approach from Jaffar's
Unification we could get rid of the map based
realization from previous releases for union
find. It turns out this gives quite a speed
advantage. We also lifted the native stack
limitation by using an extra space in the
form of a stack and a log, which surprisingly
performs well especially for Java, less for
JavaScript and Python.

- Enhanced Unary-Ops:
For unary operations such as term_variables/2,
ground/1, etc.. we experimented with both
Peter Deutsch and a stack / log approaches.
Interestingly for performance reasons we had
to dismiss the two phase approach induced by
a marking algorithm such as Peter Deutsch,
and went also with the one phase approach as
offered by a stack / log realization.

Have Fun!

Jan Burse, 02.10.2025, https://www.herbrand.ai/

--- Synchronet 3.21a-Linux NewsLink 1.2