Re: [Tzdist] [tz] Timezone in Brazil

[Paul Eggert <[e-mail]>]

On 10/12/2015 01:51 AM, Lester Caine wrote:
> you can't
> use tzdb for reliable conversions prior to 1970!

Not so. Absolutely you can use tzdata, for places like Europe/London in 
1915 where the information has been checked and double-checked. True, 
tzdata is unrelable for *some* locations before 1970, but it's also 
unreliable for some locations *after* 1970. If you want guaranteed 
reliability, you're not going to find it in tzdata, nor in any other 
database out there.

Again, the problem here is that you appear to be asking for tzdata to be 
something that it's not, e.g., to be divided into "reliable" and 
"unreliable" sections. It's not divided that way. It's divided into 
"in-scope" (most data) and "out-of-scope" (backzone) sections. This 
approach is easy to implement, because it's easy to test whether data 
entries are in scope. Dividing it into "reliable" and "unreliable" 
sections would be much harder to implement, since it would require some 
sort of measurement of "reliability" (a notion whose very definition 
would be controversial), together with an agreed-on threshold for what 
is "reliable enough". I'm not saying it would be impossible, just hard 
to do; and nobody is volunteering to do it. Certainly neither tzdata nor 
tzdist supports this sort of notion now.

> In my own case I need some way to
> ensure that backzone IS in use by clients

You can give them a copy of backzone, then, and have them compute with 
it. Or if you're using tzdist, do something similar with the servers.

There's no guarantee that clients and/or servers will have any 
particular version of tzdata. If you want a guarantee like that, the 
easiest way to implement it is to give the clients or servers exactly 
the data you want them to use, and have them use that.