Wt

Thanks for all of the work with upstream on this. I was wondering if the issue could also impact double_to_s and real_to_s in src/Wt/Dbo/backend/Postgres.C. I haven't had the opportunity to test with Boost 1.76+ yet.

It may affect those functions as well, indeed.

The bug occurs when it has to round the input up in order to fit into the given precision, so in our round_js_str we get a double as input, but ask it to output it with at most 7 decimals.

Our real_to_s function on the other hand gets a float as input, and thus already has less precision, so at the very least I can't reproduce the issue with the numbers I used to test round_js_str. That suggests we may also be able to work around the issue by sacrificing precision beforehand, or increasing precision where we use round_js_str.

If the bug affects Postgres.C as well, we're a lot less likely to trigger it. I will look further to see if there is any number that will trigger the bug.

I tried to start from 9.999f and using nextafterf to near 10.f, and didn't notice the bug.

I did encounter another issue where stof would throw an exception, though.