On Tue, Oct 6, 2026 at 6:43 PM Alisdair Meredith <alisdairm@me.com> wrote:

I remember the distinction between trigraphs and digraphs now.

Trigraphs were processed at a per-character level, essentially a form of character
encoding in source. When trigraphs were removed, IBM are able to continue
honoring trigraphs with their original meaning by making them part of the phase 1
source-to-basic-character-set mapping.

Digraphs work at the token level, so are *not* appropriate for handling as part of
the character mapping in phase 1. This shows up in things like string literals. Note
that the old rules for backing out trigraph expansion in raw string literals no longer
easily apply in the IBM case, but they seemed ready to accept that as long that they
can portably pragma the code page. Hubert might well correct me on “accept” though ;)

We still back them out in raw string literals. The world has not changed enough that simply enabling trigraphs is causing any issues yet.

As mentioned elsewhere in the thread, the "encoding" is allowed to be C++ tokenization-aware :)

There probably will come a time when we will need to turn off trigraphs (but still allow usage of ??= in place of # for introducing a preprocessor directive).

That time will probably come when some frequently re-built open source project won't accept "?\?!" in place of "??!" in some precious string literal.

-- HT