[SR-8681] lldb should use the better algorithm for float printing (like Swift)

[jepers]

Previous ID	SR-8681
Radar	None
Original Reporter	@jepers
Type	Improvement

Additional Detail from JIRA

Votes	0
Component/s	LLDB for Swift
Labels	Improvement
Assignee	None
Priority	Medium

md5: 8dd91e31acdf34ed560b155632235ce2

Issue Description:

Filing this as suggested by @jckarter in this forum post.

Swift 4.2 uses a new algorithm that should always print a minimal-length decimal value that parses back into the original floating-point value, so Double(String( x )) always produces x.

I noticed that the REPL still uses the old algorithm:

$ swift
Welcome to Apple Swift version 4.2 (swiftlang-1000.0.36 clang-1000.0.4). Type :help for assistance.
  1> 0.3
$R0: Double = 0.29999999999999999
  2> 0.3.debugDescription
$R1: String = "0.3"

Just for completion, here is the same session with Swift 4.1, where the old algorithm was used not only for the REPL's feedback but also for eg the result of 0.3.debugDescription:

$ swift
Welcome to Apple Swift version 4.1.2 (swiftlang-902.0.54 clang-902.0.39.2). Type :help for assistance.
  1> 0.3
$R0: Double = 0.29999999999999999
  2> 0.3.debugDescription
$R1: String = "0.29999999999999999"

[belkadan]

Arguably useful for LLDB in non-Swift modes too. cc jingham@apple.com (JIRA User)

[swift-ci]

Comment by Jim Ingham (JIRA)

lldb uses APFloat API's to print doubles and floats. Does the swift printing use that or something layered on top of that? If the former it should be pretty easy to include whatever flags you use to get this behavior out of APFloat. If the latter, we should move this code somewhere in llvm so lldb doesn't have to reimplement it.

[belkadan]

@stephentyrone, is the Swift logic implemented in Swift or C++?

[stephentyrone]

It's actually in C, compiled as C++, with some very lightweight Swift wrappers. We should move APFloat to use it for better performance anyway.

Semi-tangential, what would need to happen for LLDB to use debugDescription? Isn't this what it's for?

[belkadan]

debugDescription is used for 'print-object', but normal LLDB printing must not run any code in-process.