It's misleading to call this the reason for minor GC. This is the reason for GC, which may be a major or a minor (it's a major when reason & GPR_FLAG_MAJOR_MASK, and a minor otherwise).

You mean for stress specifically?

To be honest, I'm not sure exposing the stress count is very helpful, same with force, we could probably just not expose these.

I simply exposed them all without thinking too much.

No I mean for all of these. These flags tells us how the GC was started, but not what type of GC (major vs. minor). So I don't think it's correct to prefix these with minor_gc_.

Oh, I see. I had no idea. So all of these reasons above could trigger a major?

I think so.

Hum, I just dug into this, and I'm not sure you're correct:

    /* major reason */
    GPR_FLAG_MAJOR_BY_NOFREE    = 0x001,
    GPR_FLAG_MAJOR_BY_OLDGEN    = 0x002,
    GPR_FLAG_MAJOR_BY_SHADY     = 0x004,
    GPR_FLAG_MAJOR_BY_FORCE     = 0x008,
#if RGENGC_ESTIMATE_OLDMALLOC
    GPR_FLAG_MAJOR_BY_OLDMALLOC = 0x020,
#endif
    GPR_FLAG_MAJOR_MASK         = 0x0ff,

    /* gc reason */
    GPR_FLAG_NEWOBJ             = 0x100,
    GPR_FLAG_MALLOC             = 0x200,
    GPR_FLAG_METHOD             = 0x400,
    GPR_FLAG_CAPI               = 0x800,
    GPR_FLAG_STRESS            = 0x1000,

So none of these are ever gonna match the MAJOR_GC_MASK. It is true that they might be flipped up at the same time than a major reason, hence a major will trigger, but as far as I can tell they are never the reason for a major.

They aren't. They don't tell what kind of GC is being run. But that doesn't mean they guarantee a minor. The GC reason could cause both either a major or minor GC, so I don't think it's accurate to call it minor_gc_*_count.

Ok, I guess that's where the misunderstanding is.

The point isn't to say a minor was triggered. I don't expect the sum of minor_reason to match minor_gc_count.

The goal here is to see how many time we reached these limits, if a major was triggered instead I still want these to be bumped because if you change the settings to avoid the major, you'd see that minor.

Does that make sense?

Sorry, I'm getting more confused.

The point isn't to say a minor was triggered.

If a minor wasn't triggered (and a major was triggered instead), isn't it misleading to call it minor_gc_newobj_count or minor_gc_malloc_count?