On 26/05/2009, at 5:41 AM, Tom Lane wrote:
>
> The only place I can find where an oversize time value behaves in a
> seriously bogus fashion is in time_out, or more specifically
> EncodeTimeOnly(): it fails to initialize its output string at all.
> So you could easily get garbage text output, though in my quick tests
> you seem to usually get an empty string instead. The odds of an
> actual
> crash seem pretty small, but not quite zero (if somehow there was no
> zero byte up to the end of the stack).
I'm seeing all sorts of odd stuff - typically the last column value
output, but occasionally other snippets of random data that don't seem
related to the query.
> My feeling is that the error check in EncodeTimeOnly is just stupid
> and
> should be removed. That code will work fine with oversize times (and
> no, it won't overrun the output buffers either). The callers aren't
> bothering to check for error returns anyway...
I'm not sure it's postgresql's job to police things like this, but
returning values greater than 24 hours may violate assumptions in user
code, and I would be worried about potentially causing silent
failures. Of course, it should no longer be possible to get an illegal
value into the database, so the risk is low - either a database that
predates the fix, or database corruption.
Are there any other cases where the binary receive functions are
missing sanity checks?