Add setTracefile() method for structured optimization progress loggingAdd settracefile api

[MySweetEden]

Summary

Adds setTracefile() method for structured, machine-readable optimization progress logging.

Closes #1147

Motivation

As discussed in #1147, structured progress logging is useful for:

• Visualization of optimization progress
• Fair comparison of solver runs
• Analysis of primal/dual bound evolution

This provides a simpler alternative to implementing custom event handlers.

Design Decisions

• JSONL format: Allows streaming writes and partial reads even if optimization crashes
• Separate from setLogfile(): Human-readable log and machine-readable trace serve different purposes
• Append mode by default: Multiple optimize() calls can accumulate in one file
• Automatic close: Trace file is managed internally, no explicit close needed
• Field names match SCIP log: primalbound, dualbound, time, nodes for consistency

Changes

• Added setTracefile(path, mode="a") method to Model
• Internal _write_trace_event() for centralized trace writing
• Event handler for BESTSOLFOUND events

Events Recorded

• solution_update: when a new best solution is found
• solve_finish: when optimization terminates

Fields

type, time, primalbound, dualbound, gap, nodes, nsol

Usage

model.setTracefile("trace.jsonl")
model.optimize()

trace.jsonl contains JSONL records

Future Work

• Recipe/tutorial documentation will be added in a follow-up PR if this implementation is accepted

Open Questions

• Should a solve_start event be added to distinguish multiple optimize() calls in append mode?
• Output format: JSONL is chosen for streaming/crash-safety, but CSV/TSV could be discussed

[Joao-Dionisio]

Hey @MySweetEden , thank you for the contribution! However, can you please make it as a recipe, rather than a part of scip.pxi? We reserve the latter for things that are a part of core-SCIP, and we separate the helpers. So, the flow would be something like

from pyscipopt import Model
from pyscipopt.recipes.traceback import attach_traceback_eventhdlr

m = Model()
attach_traceback_eventhdlr(m)

A bit like the primal_dual_evolution recipe.

How do you feel about this approach? Everything could be implemented in Python, as well.