Re: Format base - Code contribution

Hi Chris, thanks for the reply.

On Wed, Apr 25, 2018 at 8:03 PM, Craig Ringer <craig(at)2ndquadrant(dot)com> wrote:

> Personally, I think this is a better candidate for being incorporated
> directly rather than as a contrib. This sort of utility is much less
> useful if you cannot rely on it being present.
>

I guess I've gotten used to the idea of contrib being both a test bed of
newer functionality—e.g., tsearch—without committing to a final API in the
core. Also can't imagine using PostgreSQL without pgcrypto being available.
But this is a perception issue on my part. I'm looking into where to put
this into core now.

> I'm not convinced by the wisdom of adding int8 overloads, etc, with a
> second argument. I'd rather this be named as a separate function. I
> realise that many programming languages do this, but it's IMO less
> discoverable this way, and might make our life harder if we later need
> to overload these functions in a different way.
>

Totally fair observation. Easier for users in the short term, may be harder
in the long term.

> We already have to_hex. So to_base seems a reasonable choice. Then
> adding a from_hex, from_base seems natural.

I have some misgivings about the existing to_hex now that I've had a chance
to go over it. It follows the printf model with %x for integers. I feel
this was a mistake. Hexadecimal, while enormously useful for bitwise
analysis, is still an output for human eyes. The fact that a negative int
value could be substantially different from a negative bigint value is
problematic. I understand the underlying reason for it, but a cursory check
in the mailing list archives shows more than a couple folks who got tripped
up by it.

I do not think that base 10 output should be wildly different from base 16
(or base 8). I don't think anyone would consider it intuitive to print out,
for example, 2147483647 for to_base(-1, 10), yet that's exactly what's done
for base 16 with the current implementation of to_hex. I see these problems
as apples and oranges. To be more precise, I consider the current to_hex to
be wrong, but too late to fix. to_bitwise_hex, to_raw_hex, or similar would
be more appropriate. In C, it's clear at all times what the size may be.
Within an SQL query, things can become far more ambiguous.

Most modern, high-level languages will present 15 as hex F and -15 as hex
-F, which is uniform no matter the underlying type size. All numeric types
in PostgreSQL are signed. Getting a wildly different value because some
smallint got silently coverted into an integer is non-intuitive to say the
least.

So it would appear there should be a strict demarcation between to_hex and
the proposed to_base.

> Bonus points if you add
> to/from base64 and oct while you're at it.
>

I can happily do it, but again, I think from_hex and from_oct should follow
as inverses to_hex and to_oct, not to_base/from_base for the reasons given
above. As for base64, that's another problematic one. To most folks, base64
means a binary encoding of data into ASCII. Again, solving a different
problem. I think it would be a good idea to avoid mixed messages to the
user here even if to the point of limiting to an upper limit of base 62
(0-9, A-Z, and a-z) and erroring out above that. I'd like to go to 64 if
for no other reason than the power of 2 affinity, but I don't think it
should be done lightly at the expense of user confusion. On the bright
side, encode/decode are both well-established within PostgreSQL and clearly
dealing with bytea values rather than integer values.

We don't seem to have a "from_hex" or "int8_from_hex", which is a
> bewildering oversight really, and we don't accept literals:
>

Thanks for the illustration into PostgreSQL parser behavior. The
flexibility of PostgreSQL can obviously be both a curse and a blessing.
Hoping I can add to the blessings.

--
Quidquid latine dictum sit, altum sonatur.
- Whatever is said in Latin sounds profound.