Hacker News Reader

Comment by graemep | Hacker News Reader

> Early versions of SQLite (prior to 2004) operated on the classic TCL principal that "everything is a string". Beginning with SQLite3 (2004-06-18), SQLite also supports binary data.

TCL can handle binary data. It is just not a separate type: https://wiki.tcl-lang.org/page/Working+with+binary+data

SQLite also always had a null type, surely?

> However, types are still very flexible in SQLite, just as they are in TCL. SQLite treats the datatypes on column names in a CREATE TABLE statement as suggestions rather than hard requirements

This is something I do not much like. Its not compulsory (you can create "strict" tables). It works well in TCL which is an entire language designed around the idea. Less so in SQL.

One of the advantages of RDBMSes is that not accepting obviously wrong data makes life easier for developers. You can debug an issue that happens on inserting the data, not when you find the wrong type or other bad much later on.

View on HN

There was a time when TCL could not handle binary data at all. The support came in TCL 8.0:

"Binary data is now supported in Tcl."

https://www.tcl.tk/software/tcltk/relnotes/tcl8.0.txt

Well before 2004, but worth mentioning because you'll find a fair amount of old posts complaining about it.

What do you mean, that TCL strings weren’t 8-bit-clean?

Maybe they were zero-terminated C strings, which of course can’t represent arbitrary binary data.

Yes, prior to 8.0, tcl strings mapped to null-terminated C strings.

"Binary I/O. The new object system in Tcl 8.0 supports binary strings (internally, strings are counted in addition to being null terminated). There is a new "binary" command for inserting and extracting data to/from binary strings. Commands such as "puts", "gets", and "read" commands now operate correctly on binary data. There is a new variable tcl_platform(byteOrder) to identify the native byte order for the current host."

This typing behaviour, now in combination with strict tables, is a boon for biostats. When you get shitty data to be cleaned, you've got 3 main choices: 1.use slow and untyped scripting languages, 2.use a strictly typed database, meaning you'll have to clean your data in advance, or 3.load it all as strings into SQLlite, then clean the data until it fits into a strict table with check constraints. IMO, it's pretty clear 3 is best by far!

I agree that 3 is great, but you can also do that in any database, just create your input tables as string only and then perform the necessary operations to move them into typed tables.

Yes, but with sqlite there is much less ceremony (no server, etc.) and most importantly can be used without talking to my institution's sysadmin, which is what I'm looking for when manipulating one-off datasets.

But that advantage has nothing to do with accepting data of the wrong type into a column (by default).

Is anyone using untyped languages much today, other than shell scripts?

CMake is entirely stringly-typed as well. Like many shells, arrays/lists are just space-separated strings.

Pedantically that's not true, they're ';' delimited https://cmake.org/cmake/help/v3.31/command/list.html#:~:text...

The confusion comes from the fact that set() automatically coerces space-delimited items into a ;-delimited list

  set(ONE alpha;beta)
  set(TWO alpha beta)
  list(LENGTH ONE one_len)
  list(LENGTH TWO two_len)
  message(FATAL_ERROR "one <<${ONE}>> length ${one_len}\ntwo <<${TWO}>> length ${two_len}")

emits

  one <<alpha;beta>> length 2

  two <<alpha;beta>> length 2

Most people via JavaScript...

> SQLite also always had a null type, surely?

In 'stringly typed' languages, the natural value of NULL is the empty string.

SQLite currently distinguishes between "" and NULL, and I don't know if this was always the case. But it's quite possible that SQLite 1 and 2 conflated the empty string and NULL.

> SQLite also always had a null type, surely?

NULL is surely a value, not a type, no? You can restrict use of NULL with the column type, but it doesn't make any sense to have a NULL type rather than, say, a nullable-string type. What would be the point?

Null/none is a value and a type. E.g. in Python there's None which has the type NoneType. Sometimes they are named the same.

This is an example of a singleton type - a type that can only hold one value. It's not the only example though, it's fairly common to support string or integer singleton types, e.g. in Typescript you can have the type "foo" which only has a single possible value, "foo". Or some languages support a type int(5) for which the only possible value is 5.

You might think that's useless, but it's very useful when combining types, e.g. via unions. Or sometimes when writing generic code.

SQL has an extremely primitive type system so there's no syntax for type unions etc. But you can imagine if it was written by someone who had experience of modern typing you would say `my_nullable_column: string | null`.

Ok, but you could also just say "nullable string". Giving null a type doesn't seem to yield a benefit outside of python. Especially when, again, it doesn't appear to have any utility in SQL outside of applying to types.

Reply on news.ycombinator.com