| Age | Commit message (Collapse) | Author |
|---|
|
Arrange for all the variable declarations in indent_globs.h to look
like "extern" declarations to every .c file except indent.c.
This prevents linker failure due to duplicated global variables when
the code is built with -fno-common, which is soon to be gcc's default.
The method of temporarily #define'ing "extern" to empty is a hack,
no doubt, but it avoids requiring a duplicate set of variable
definitions, so it seemed like the best way.
Discussion: https://postgr.es/m/20200629165051.xlfqhstajf6ynxyv@alap3.anarazel.de
|
|
Previously, we would assume that a macro like IsA() in the following
example was a cast just because it mentions a known type between parens,
and that messed up the formatting of any binary operator that followed:
if (IsA(outer_path, UniquePath) ||path->skip_mark_restore)
This change errs on the side of assuming that function-like macros are
similar to sizeof() and offsetof(), so that operators are formatted
correctly:
if (IsA(outer_path, UniquePath) || path->skip_mark_restore)
Thomas Munro
Discussion: https://postgr.es/m/20200114221814.GA19630@alvherre.pgsql
|
|
Since the recommended build method uses Postgres' preferred CFLAGS,
and those now include -Wimplicit-fallthrough, we now see these
warnings where we did not before.
Michael Paquier
Discussion: https://postgr.es/m/20200515060318.GB79590@paquier.xyz
|
|
The upstream maintainer doesn't want these, so drop 'em.
Also, teach is_func_definition() about //-style comments. The main
lexi.c code doesn't know about those either, but I assume somebody
will wish to fix that at some point.
|
|
The previous coding for deciding whether "foo(" begins a function
definition or just a declaration was quite stupid: it could be
fooled by nested parentheses in the parameter list, or comments
in or after the parameter list, or whitespace just after the
parameter list. And it didn't work if the parameter list crossed
a line boundary, either.
We can upgrade all those cases without any objectionable side
effects, I believe. However, there's one more case that's
problematic: if there's non-punctuation stuff after the
parenthesized arguments, it's difficult to tell whether that
stuff is K&R-style parameter declarations or function attributes.
I don't care to try to make the lookahead code smart enough to
tell the difference, so instead let's invent a switch to tell
what to do. The '-kr' mode is backward-compatible, hence the
default. The '-nkr' mode will do the right thing with function
attributes and the wrong thing with pre-ANSI function declarations;
but for a code base with none of the latter, that hardly matters.
|
|
This is a workaround for an acknowledged upstream bug; it basically
reverts this logic to the way it was in NetBSD indent. While there
may eventually be a different upstream fix, this will do for our
purposes. Add a regression test case that illustrates the problem,
and tweak declarations.0.stdout for slightly different formatting
produced for a function pointer typedef.
|
|
This switch enables using Postgres' rules for using tabs vs spaces
in indents, to match the output of our old "entab" program.
The FreeBSD maintainer wants no part of this, so we'll have to carry
this as a forked patch. (The alternative is to go over to use-tabs-
whenever-possible rules, but that would result in circa 10K lines of
invisible whitespace diffs in the PG sources, which seems like lots
more pain than is justified.)
|
|
Remove '#include <sys/cdefs.h>', which will fail in many non-BSD
environments, as well as the __FBSDID() FreeBSD-ism. Replace them
with '#include "c.h"', which will allow use of the Postgres
project's portability infrastructure. This is intended to be a
permanent diff between the FreeBSD upstream files and our copy.
|
|
This commit copies Piotr Stefaniak's development repo
https://github.com/pstef/freebsd_indent
as of commit 5d3b621cd54de672983c51d9622dee84d458dd83 (2017-06-18).
But I've omitted Makefile and Makefile.depend, as they are of no use
in non-BSD build environments.
|
