Don't overflow the bit buffer printing Cd data in binary ##crash - #27322

• Mark this if you consider it ready to merge
• I've added tests (optional)
• I wrote some lines in the book (optional)

Description

With a binary immbase (ahi 1), ds_print_data_type prints a Cd item bit by bit with r_str_bits (msg, buf, size * 8, NULL). msg was char msg[64], and buf only holds the first 8 bytes of the item:

• for an 8-byte item, the 64 bits plus the NUL wrote one byte past msg;
• for a 31-byte item (.smi), it wrote 249 bytes into msg and read 31 bytes from the 8-byte buf. wx 0102030405060708; Cd 31; ahi 1; pd 1 aborts with *** stack smashing detected ***.

msg now holds 64 bits plus the NUL. Items larger than 8 bytes fall back to the hexdump, which is what pd already shows for Cd 31 without the hint.

#27320 moves this code into a helper. I'll rebase whichever of the two lands second.

Testing

• Two tests in test/db/cmd/metadata:

  • an 8-byte item in binary, printing all 64 bits;
  • a 31-byte item, falling back to the hexdump. It crashes on master.

  The 8-byte test passes on master too, because a build without ASan does not notice the one extra byte.