lihaoyi
commented
9 months ago Closed mokexinxin closed 9 months ago
lihaoyi
commented
9 months ago lihaoyi
commented
9 months ago Sjsonnet works with a Javascript data model. That means double precision floating point numbers. There really isn't anything we can do here: Even if uPickle is able to read the raw JSON number as a full un-truncated string, Sjsonnet does not, and
It's unlikely we can use a different number representation for Sjsonnet, due to performance considerations: BigDecimals are orders of magnitude slower and more memory-intensive than Doubles, and Sjsonnet is somewhat performance sensitive (perf is it's raison d'etre for existing)
databricks/sjsonnet is meant to be equivalent to google/jsonnet, and google/jsonnet also truncates long numbers. It truncates your example differently - which is its own issue I've opened as https://github.com/databricks/sjsonnet/issues/186 - but it still truncates it, so your use case is not supported on any implementation of the Jsonnet language
You'll have to wrap your long numbers it in a string, or use a different configuration language that uses arbitrary precision numbers
Hello: I want to use JSON to read the 18 digit numbers generated by the snowflake algorithm, but I encountered an accuracy loss issue after reading, as follows
{'a':738529527931269425}But after I ran it, I showed the following results:{"a":738529527931269376}The running result shows incorrect numbers, missing is accuracy. If I want to fully obtain this number without converting it to a string, is there a solution?
Note: This issue has been encountered while running sjsonnet. Preliminary troubleshooting may be in this project, and we hope the author can help solve it. Thank you very much!