core: Accept a b/k/m suffix in size-valued directive parameters

[notroj]

ap_size_parameter_parse() now accepts an optional case-insensitive b, k or m suffix (bytes, ×1024, ×1024×1024) after the number, rejecting overflow with APR_ERANGE. Size-valued directives using it, such as LimitRequestBody, ScriptLogLength, ScriptLogBuffer, DavQuota, BufferSize and AuthFormSize, thereby take the same suffixes AuthDigestShmemSize already did.

A second commit converts mod_auth_digest's set_shmem_size() to the shared parser. Parsing is strict: anything after the suffix is rejected, including trailing whitespace (only reachable via a quoted argument), which set_shmem_size() previously tolerated.

Unit tests added in test/unit/util.c.

🤖 Generated with Claude Code

[arturobernalg]

I think the overflow checks need to use the range of apr_off_t, rather than APR_INT64_{MIN,MAX}. On a platform with 32-bit apr_off_t, e.g. 2097152k passes the current check and then overflow in parsed *= 1024. The "4G" success test has the same assumption.
Also, this will want #848 (or equivalent) before merge, otherwise the new no-conversion tests still differ on musl/FreeBSD.

[notroj]

I think the overflow checks need to use the range of apr_off_t, rather than APR_INT64_{MIN,MAX}. On a platform with 32-bit apr_off_t, e.g. 2097152k passes the current check and then overflow in parsed *= 1024. The "4G" success test has the same assumption. Also, this will want #848 (or equivalent) before merge, otherwise the new no-conversion tests still differ on musl/FreeBSD.

Makes sense, thanks for the review! Will do