Dumping should fail for lists with mixed data types

[metatoaster]

Currently, the behavior of dumps on data that contain a list of mixed types will produce invalid toml serialization successfully instead of raising an exception. For example:

>>> import toml
>>> toml.dumps({'a': {'b': ["a", [1, 2]]}})  # b mixes string with array
'[a]\nb = [ "a", [ 1, 2,],]\n'

Which naturally result in an invalid toml document that can't be loaded back:

>>> toml.loads(toml.dumps({'a': {'b': ["a", [1, 2]]}}))
Traceback (most recent call last):
 ...
IndexError: list index out of range

Compared to the node implementation @iarna/toml which has a more compliant behavior on the same input:

> const TOML = require('@iarna/toml')
> TOML.stringify({'a': {'b': ["a", [1, 2]]}})
Error: Array values can't have mixed types

[eksortso]

The TOML standard currently does not permit mixed-type arrays. To be compliant with TOML v0 5.0, yes, heterogenous arrays must throw an exception.

It may be worth noting there's talk of allowing mixed types in arrays in future versions of the standard. It appears to be gaining traction, because of kludgey ways of working around the restriction and TOML's drift away from just being a configuration format.

So far, allowing mixed types has not been approved. But for this project and others, I would recommend maintaining logic allowing for mixed types, even if it isn't always used. Depending on how the standard evolves, the current non-compliant behavior may still be useful.

[metatoaster]

The problem I clearly demonstrated here is that the loader cannot round-trip (de)serialization of a particular input; regardless of current/future specification conformity this is a problem that can manifest as bugs for downstream packages that accept user-input that somehow successfully serializes but could not then reloaded. As it is now this also forces packages depending on toml to apply additional logic that should have been provided by toml to ensure input passed to toml.dumps will produce an output that can be successfully loaded again by toml.loads. If in the future mixed types is allowed it would simply disable whatever additional checks that should be implemented inside toml.dumps.

[ghost]

It's not really an actual problem anymore, because with TOML version 1.0.0 values of different types may be mixed.
Doc (https://toml.io/en/v1.0.0#array).

Now different types can't be mixed, this problem is already illustrated here: #270

[pradyunsg]

Indeed!

I'll close this out in that case. :)