Are empty tables prohibited or not?

[westluke]

Shriram and I noticed a funny thing about tables the other day!

If you try to create a table with an empty schema using table: end, you get the message "This table has no column names, but tables must have at least one column." But that's not true! Because the expression (table: colname row: 3 end).drop("colname") doesn't throw an error, and gives you a table without columns.

The B2T2 spec requires the ability to make tables with an empty schema, so it seems like the table: end syntax should work.

Originally posted in the code.pyret.org repo, but that was a mistake.

[bennn]

My 2c:

• table: end should be an error (but reword the message be be true), and
• there should be a special constructor (empty-table() ?) for saying yes, I really do want an empty table

[jpolitz]

I think @bennn 's suggestion is great. The error can say “if you really meant to make an empty table, use empty-table”

[shriram]

I like it too, with that error message!

[blerner]

Hmm, counterargument: table: x, y, z end is valid. table: x, y end is valid. table: x end is valid. So shouldn't table: end be valid and be a zero-column, zero-row table?

[shriram]

Well, that was the original argument. Why was it ever disallowed?

[blerner]

I think it was a question of whether the table could grow or not. An empty table with multiple columns could add new rows. Could an empty table with no columns add new rows? Explicitly, is [raw-row: ] a valid value?

[shriram]

Yes, it would seem to me we can have a table of no columns with N rows. Then, when we add a column, we can provide (say) a list of N values and each one would extend one of those empty rows.