Re: Pass heap tuple counts to parallel vacuum as double

Hi,

On Mon, Nov 9, 2026 at 10:03 PM Chao Li <li(dot)evan(dot)chao(at)gmail(dot)com> wrote:
>
> The patch also looks good to me.

Yes, it LGTM too.

"long" is a signed 8 bytes everywhere except on Windows, where it is 4
bytes, and passing a number of tuples more than the 4-byte limit could
potentially expose it to overflows.

-Wfloat-conversion tells me there are other such uses [1] (haven't
checked in depth whether these are worth fixing, and even if they need
to be, in a separate thread of course).

> I wonder if we can still get this into 19, as it hasn’t been released yet. This may be our last chance to change these function signatures for 19.

+1 to back-patch to PG19 (assuming we have another release after RC2
before the planned GA on Nov 19th) to avoid the overflow issues and
keep the version diff small.

[1] gininsert.c:1689:4: warning: conversion to 'long int' from
'double' may alter its value [-Wfloat-conversion]
vacuumlazy.c:2568:52: warning: conversion to 'long int' from 'double'
may alter its value [-Wfloat-conversion]
vacuumlazy.c:3012:52: warning: conversion to 'long int' from 'double'
may alter its value [-Wfloat-conversion]
nbtsort.c:499:4: warning: conversion to 'long int' from 'double' may
alter its value [-Wfloat-conversion]
vacuum.c:2568:13: warning: conversion to 'long int' from 'double' may
alter its value [-Wfloat-conversion]
variable.c:333:16: warning: conversion to 'long int' from 'double' may
alter its value [-Wfloat-conversion]

--
Bharath Rupireddy
Amazon Web Services: https://aws.amazon.com