Export fails for dates before 1980

[stefan-sherwood]

To report a bug...

• What command(s) did you run?

task add description:"Test date" modified:"1970-01-01T00:00:00Z"

If I do the identical command with a slightly different date in the same format, it correctly adds the task with the given time/date,

$ task add description:"Test date" modified:"1970-01-01T00:00:01Z"
Created task 44.

Note that the date format I gave is correct, both according to ISO-8601 rules and according to taskwarrior specification for supported dates.

• What did you expect to happen?

Task to be added.

• What actually happened?

Error message: '1970-01-01T00:00:00Z' is not a valid date in the 'Y-M-D' format.

Task was not added.

• Paste the output of the task diag command.

task 3.0.2
   Platform: Linux

Compiler
    Version: 13.2.0
       Caps: +stdc +stdc_hosted +LP64 +c8 +i32 +l64 +vp64 +time_t64
 Compliance: C++17

Build Features
      CMake: 3.27.4
    libuuid: libuuid + uuid_unparse_lower
 Build type: release

Configuration
       File: /home/stefan/.taskrc (found), 1381 bytes, mode 100644
       Data: /home/stefan/.task (found), dir, mode 40755
    Locking: Enabled
         GC: Enabled
Hooks
     System: Enabled
   Location: /home/stefan/.task/hooks
             (-none-)

Tests
   Terminal: 350x113
       Dups: Scanned 51 tasks for duplicate UUIDs:
             No duplicates found
 Broken ref: Scanned 51 tasks for broken references:
             No broken references found

[djmitche]

I suspect this is because zero is treated as a sentinel value to indicate that the time wasn't converted, and the string ending in :00Z evaluates to zero as a UNIX timestamp.

[stefan-sherwood]

That makes sense. I am trying to run a script from the syncall project that synchronizes Google Keep with taskwarrior. It uses a zero timestamp when there is no modification date, presumably to ensure that all other modification dates compare as being later.

If this is considered functioning as designed, I understand.

As an aside it's a bit strange that it interprets the format was being 'Y-M-D' when it pattern matches an ISO-8601 date/time format but that's just a nitpick about the error message.

[djmitche]

It's trying to interpret that using the dateformat config, but I think that allows some other, fixed formats as well (including all of the relative times like tomorrow or soq and apparently ISO-8601 times)

I think relying on 0 as a sentinel value was based on the idea that nobody was using Taskwarrior in 1970, but the truth is probably lost to time. Presumably we could make 0 valid and return an exception. But I don't know much about the "variant" and "evaluation" stuff in src/columns/ColTypeDate.cpp to know how to do that.

The error message could use some improvement, for sure. Maybe if that just said "Could not parse .. as a date after 1970" it would at least be accurate? WDYT?

[stefan-sherwood]

Looping in the author of syncall @bergercookie to the conversation since this is really about getting the two projects to agree on what makes sense in this context. Having a value that represents "the beginning of time" makes sense to me and having that value be 0 seems to be the right thing to do. Obviously I could change it to pass in 1 but that seems like a hacky workaround.

IMO taskwarrior clients should not be subject to the effects of project-internal sentinel values. I'm not at all familiar with the code but in general if a Variant needs to be able to have a type before its value is set then an additional boolean (e.g. bool _isset) to indicate there's a value is merited. Otherwise, there could be an additional type
(e.g. type_none) added that indicates it doesn't have a type.

In terms of the error message, my primary complaint is that it sounds like the reason it's not working is because it's comparing against a different format (YYYY-MM-DD) when ISO-8601 actually is accepted. The comment on the code implies the same thing. The simplest change to change the error message to:

'1970-01-01T00:00:00Z' is not a valid date. See https://taskwarrior.org/docs/dates for more info.

And also update the documentation to indicate what date ranges are accepted since ISO-8601 itself has no such restriction.

As another side note, the syncall project actually uses the now-deprecated truncated ISO-8601 format. And taskwarrior accepts it, even though it's not listed in the taskwarrior documentation as accepted.

[djmitche]

Hopefully someone more familiar with the Variant stuff can fix the internal sentinel values.

But the change to the error message and the two docs fixes sound like great PR(s)!

[stefan-sherwood]

I'm happy to do that but I cannot find where the documentation at https://taskwarrior.org/docs/dates/ is located to edit.

[djmitche]

It's in the https://github.com/GothenburgBitFactory/tw.org repo.

…

On Wed, Jul 31, 2024 at 7:28 PM stefan-sherwood ***@***.***> wrote: I'm happy to do that but I cannot find where the documentation at https://taskwarrior.org/docs/dates/ is located to edit. — Reply to this email directly, view it on GitHub <#3572 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAHAAJ32F3BN25GINN5VKDZPFXJHAVCNFSM6AAAAABLW3MTMCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRRGY2TGMBVHA> . You are receiving this because you commented.Message ID: ***@***.***>

[tbabej]

@stefan-sherwood did we end up changing the docs here?

[tbabej]

The problem with using beginning of unix time as an actual value reproduced on 3.1.0.