Fix using optional between numbers
#362
Conversation
|
There is still a way to make parsers fail when their specification is unambiguous, for example, with something like as the trie doesn't get merged with the neighboring number consumers. Fixing it is completely doable, which is why we don't restrict this usage, but it would severely overcomplicate the code and would probably not help anyone at all. |
| @@ -129,6 +129,8 @@ public sealed interface DateTimeFormatBuilder { | |||
| * | |||
| * [am] is used for the AM marker (0-11 hours), [pm] is used for the PM marker (12-23 hours). | |||
| * | |||
| * Empty strings can not be used as markers. | |||
There was a problem hiding this comment.
Do we check this requirement in the amPmMarker call, or it fails the operation (the parsing I suppose) later?
Do we need the requirement to have distinct strings here?
There was a problem hiding this comment.
The amPmMarker call checks this requirement. Documented this and added the tests.
Also, added the requirement for distinct strings for month days, days of the week, and AM/PM.
dkhalanskyjb
force-pushed
the
fix-optional-between-numbers
branch
from
07a8202 to
b430873
Compare
| @@ -21,6 +23,14 @@ public class MonthNames( | |||
| ) { | |||
| init { | |||
| require(names.size == 12) { "Month names must contain exactly 12 elements" } | |||
| names.indices.forEach { ix -> | |||
There was a problem hiding this comment.
On the second thought about duplicates, this way users would not be able to use single-letter month names for formatting.
There was a problem hiding this comment.
After an internal discussion, we decided that it's not an issue with the typical use cases for single-letter month names: they are mostly used independently from the other formatting, so you probably won't see things like 2024-J-06 in the wild. Instead, you can find a row of J, F, M, A, etc., as labels on a chart or buttons in a calendar app, but then, there is no reason to build a custom format, you can just do something like date.month.name[0].
The reproducer was:
```
offsetHours(Padding.NONE)
optional {
optional { char(':') }
offsetMinutesOfHour()
}
```
This showed us one bug and one inefficiency, both of which are
fixed here.
Inefficiency: the `optional { char(':') }` should for all intents
and purposes be equivalent to
`alternativeParsing({ char(':') }) { }`: both the empty string and
the `:` should be parsed, and, according to `optional`'s
definition, if none of the fields mentioned in the block are
non-zero, nothing should be output. However, such `optional` still
created a fake parser element that notified all the fields that
they are zero when an empty string is parsed, even though there are
no fields.
Bug: the fake parser element that notifies the fields that they
are zero even though nothing of note was parsed interfered with the
mechanism supporting compact formats. Number parsers are greedy,
so `number(hh); number(mm)` gets merged into `number(hhmm)`.
If there is something between them, this merge can't happen:
`number(hh); string("x"); number(mm)`.
For this reason, parsers that don't accept any strings are
forbidden (or just very unlikely to happen)--except for the
zero-width unconditional modification parser. This bug is fixed by
moving such parsers to the end of the parser:
`number(hh); modification(); number(mm); string("x")` will get
transformed to `number(hhmm); string("x"); modification()`.
dkhalanskyjb
force-pushed
the
fix-optional-between-numbers
branch
from
b430873 to
b5f40cd
Compare
See the commit message for details of the bug.