math/big: Support string numbers in `Int.UnmarshalJSON`

[lmittmann]

What version of Go are you using (go version)?

$ go version
go version go1.17.3

Does this issue reproduce with the latest release?

Yes

What did you do?

Try to decode json numbers in string format as big.Int.

What did you expect to see?

String number (e.g. {"val": "123"}) should be unmarshaled to big.Int.

What did you see instead?

String number (e.g. {"val": "123"}) can not be unmarshaled to big.Int.

Proposal

Even though the json spec does allow numbers of any size it is very common to have big numbers in string format, e.g. {"val": "123"} instead of {"val":123}. I propose to add support for reading numbers in string format to big.Int.UnmarshalJSON.
No API change, just a small improvement.

Related Issues:

• encoding/json: marshal of custom type not quoted when using ",string" tag field  #20651
• math/big: big.Int (incorrectly) marshals to a number instead of a string #28154
• math/big: big.Int JSON marshalling to/from string #36099

[mvdan]

Is this not a duplicate of #36099? If we were to do this, I'd think we should restrict it to ,string, to not break the behavior of existing programs. For example, a program using big.Int right now only accepts JSON integers, and not JSON strings, so we'd be making it accept an input that it didn't before. cc @dsnet

[lmittmann]

Is this not a duplicate of #36099?

#36099 also wants to enable string marshalling by adding ,string to big.Int.

If we were to do this, I'd think we should restrict it to ',string', to not break the behavior of existing programs.

You are right. I can't think of a reasonable use-case though where a string number as input should break things. But it will definitely break test suites.

Should I close this issue in favor of #36099? It seem to be stale.

[seankhliao]

Duplicate of #36099