Unary minus operator at beginning of expression is ignored about jinjava

[tkindy]

Just to check my understanding, I confirmed that Jinja handles this as I would expect.

[tkindy]

I believe I've found the bug, and it's even more basic than the examples so far in this issue. The template {{ -10 }} parses as -10 in Jinja and 10 in Jinjava.

The bug is coming from a feature of Jinjava (which, as expected, comes from Jinja) called whitespace control. In an expression or other block, you can add a dash at the beginning or end to control whether preceding or trailing whitespace on the line is trimmed. For example, the template {{- "foo" }} would trim all leading whitespace on the line ahead of foo.

The Jinja docs specify, "You must not add whitespace between the tag and the minus sign." However, the current Jinjava implementation allows whitespace in between the tag and the dash, so negative numbers at the start of expressions are having their minus sign interpreted as a whitespace control character by accident.

This is also the case in statements ({%), though I don't know of a valid statement off the top of my head that could start with a negative number.

I should be able to fix this relatively easily.

[tkindy]

I've managed to fix the whitespace control problem locally, but it's revealed another issue with operator precedence. Specifically, the unary minus should bind tighter than operators like is; that is, -1 is integer should evaluate the same as (-1) is integer. However, currently Jinjava evaluates this as -(1 is integer) which is obviously incorrect.

This is a tougher issue to untangle, but I did a fair amount of digging yesterday and hope to make some more progress today. I opened #893 to capture this issue.

[tkindy]

I had to revert #896 while we look into some templates that no longer parse successfully with these changes.