Bug 7481 – Compiler should 'soldier on' after template errors

At the start of TemplateInstance::semantic is this code: if (global.errors) { //printf("not instantiating %s due to %d errors\n", toChars(), global.errors); if (!global.gag) { /* Trying to soldier on rarely generates useful messages * at this point. */ fatal(); } It's a hack, really. This has the effect that after ANY error, any template instantiation will abort compilation completely. Back in the Bad Old Days, when the compiler soldiered on, it would spew reams of garbage and then crash. But now, with all of the ErrorExp improvements, soldiering on does in fact generate useful error messages. If there have been errors in the template arguments, obviously it shouldn't instantiate it, but there's no need to abort compilation completely. More importantly, notice that it *does* continue if errors are gagged. Aborting compilation is hampering our ability to observe what it's doing. Soldiering on will help make it more robust. Simple example: this code should generate two errors, one for 'zoo' and one for 'roo'. int foo(X)(X x) { return 6; } void main() { foo!(zoo)(5); foo!(double)(roo); }

The place I'd like to get is to continue compiling anything that does not depend on something that already failed. We're partway there with the way expressions are evaluated. For templates, it should try to instantiate one only if its arguments are free of errors.

(In reply to comment #1) > The place I'd like to get is to continue compiling anything that does not > depend on something that already failed. We're partway there with the way > expressions are evaluated. > > For templates, it should try to instantiate one only if its arguments are free > of errors. That's basically what this patch does.

I'm seeing a huge amount of error messages. It's excessive, please improve the situation.