Le Wang wrote:

> So two issue come up: should symbol even be supported by `eval-after-load`
> if the feature might not be finished loading? If it has to be kept for
> backward compatibility, documentation should explicitly say don't use
> symbol because the feature might not be finished loading.

The version of eval-after-load in the trunk handles this better, so that
even if provide is not at the end of the file, the eval-after-load forms
do not run until the file has finished loading.

What is the second issue?