Re: Recommend wrappers of PG_DETOAST_DATUM_PACKED()

On Tue, Oct 25, 2016 at 10:57:24AM -0400, Noah Misch wrote:
> When commit 3e23b68dac006e8deb0afa327e855258df8de064 introduced single-byte
> varlena headers, its fmgr.h changes presented PG_GETARG_TEXT_PP() and
> PG_GETARG_TEXT_P() as equals. Its postgres.h changes presented
> PG_DETOAST_DATUM_PACKED() and VARDATA_ANY() as the exceptional case. Let's
> firmly recommend PG_GETARG_TEXT_PP() over PG_GETARG_TEXT_P(); likewise for
> other ...PP() macros. This shaves cycles and brings consistency of style.
>
> If the attached policy-setting patch seems reasonable, I will follow it with a
> mechanical patch adopting the recommended macros throughout the tree. (If any
> code does want the alignment it gets from an obsolecent macro, I will add a
> comment to that code.)

That mechanical work unearthed special cases, hence the delay. Patch series:

1. Fix old comment about length of text, bytea, etc.

2. Unify style around deconstruct_array().

3. Fix pg_file_write() error handling. This is largely unrelated, but I
noticed it when I was about to update the first VARSIZE() in that function.
I'll back-patch this one.

4. Policy change. This is a new version of the patch I posted upthread. I
added a recommendation of PG_GETARG_INET_PP() over PG_GETARG_INET_P(). I
added updates to xfunc.sgml, src/tutorial, and a few more comments.

5. Use the now-preferred interfaces in more places. The log message describes
how I resolved some arguable points. In particular, I declined to change
areas where the detoasted varlena header leaks into longer-lived state.

Having stumbled on rijndael.c, I am not enormously confident that I noticed
all the places expecting alignment in bytea or text. This patch series does
pass "make check-world" on a system where an unaligned 2B, 4B or 8B read
elicits SIGBUS. If I did miss such an expectation, the result should be a
SIGBUS or a performance loss, not a silent malfunction.

Thanks,
nm