[iluuu1994]

/cc @henderkes

[henderkes]

Is it just because of the compile-time offset of CG or EG from _tsrm_ls_cache? Otherwise an ABI change could be an issue in NTS too, wouldn't it?

If it is limited to ZTS because of the former, perhaps we could add the padding (or set alignment) on the parent struct?

If it's the latter, this is the right approach.

[iluuu1994]

@henderkes

Is it just because of the compile-time offset of CG or EG from _tsrm_ls_cache?

Yes, if the structs grow ZEND_CG_OFFSET/ZEND_EG_OFFSET will be wrong when compiled against older PHP versions. Technically sizeof(global) itself is already an ABI break but one that we can effectively ignore, as globals aren't copied around or initialized by extensions.

If it is limited to ZTS because of the former, perhaps we could add the padding (or set alignment) on the parent struct?

I don't quite understand. You mean omit the offset for NTS? Technically the padding is not needed there, but we can also avoid the other technical ABI break above with this, so it doesn't hurt.

[henderkes]

I don't quite understand. You mean omit the offset for NTS.

I meant specifying the padding at the compile time offset define. But as this also fixes the technical ABI break in NTS, yes, this is better.

Let me check something about alignment later on desktop, other than that +1.

[iluuu1994]

Adding the padding to the offset number directly also means you'd have to sync the two locations, right? So I think this is more explicit.

Let me check something about alignment later on desktop, other than that +1.

Thanks! This will need to land tomorrow to get included in the last RC. Sorry to rush you. 😅

[henderkes]

Home in an hour and a half!

[iluuu1994]

@henderkes Great, thanks for checking