early stopping always puts objective as stopping metric, when monitoring several metrics and setting first_metric_only=True

[jc-numeus]

Description

when i include the objective function in the metric parameter list besides other metrics, even if i put another metric such as pearson at the beginning of the list, and also at the beginning at feval (which i think correctly makes it the 'first metric'), when i use first metric only = True, the early stopping always switches to stopping based on my objective function. i debugged this down to a breakpoint in the early stopping callback python implementation, this line:
self.first_metric = env.evaluation_result_list[0][1].split(" ")[-1]
here, it seems that evaluation_result_list is somehow being resorted before, such that the first metric is the objective, even though i want to stop based on for instance pearson correlation.

Reproducible example

...

Environment info

LightGBM version or commit hash: 4.1

Command(s) you used to install LightGBM

Additional Comments

[jmoralez]

Hey @jc-numeus, thanks for using LightGBM. Can you please provide a reproducible example?