On 11/09/2020 23:26, Jakub Kicinski wrote:
> "Toolchain" sounds a little grand in this context, the script that
> parses kdoc does basic regexps to convert the standard kernel macros:
> ...
> IDK if we can expect it to understand random driver's macros..
I wasn't suggesting it should _understand_ this macro, justrecognise
when something _is_ a macro it doesn't understand, and refrain from
warning about it in that case.
But I don't know how hard that would be to achieve — not being fluent
in Perl, scripts/kernel-doc is mostly line noise to me. My best
guess is that on seeing 'DECLARE_A_THING(name, argument);' it reads
it as though it were 'DECLARE_A_THING name, argument;', mistaking
the macro name for a type.
I think the only way non-macro declarations can legitimately contain
parens is in function pointer types, in which case the contents of
the parens are part of the type and not declared identifiers. So
shouldn't kernel-doc ignore anything within parens?
> This is the only use of _MCDI_DECLARE_BUF() in the tree
Well, except for mcdi.h's
#define MCDI_DECLARE_BUF(_name, _len) _MCDI_DECLARE_BUF(_name, _len) =
{{{0}}}
> , how about converting the declaration to:
>
> #declare _MCDI_BUF_LEN(_len) DIV_ROUND_UP(_len, 4)
>
> efx_dword_t txbuf[_MCDI_BUF_LEN(MC_CMD_PTP_IN_TRANSMIT_LENMAX)];
>
> Would that work?
That could work, yes. Though, probably lose the leading underscore
in that case.
-ed
