[clang] [analyzer] Model concrete floating-point values (PR #214098)

[Balázs Benics via cfe-commits]

================
@@ -0,0 +1,78 @@
+// Multiple targets needed because long double semantics differ on them.
+//
+// RUN: %clang_analyze_cc1 -triple x86_64-unknown-linux-gnu \
+// RUN:   -analyzer-checker=core,debug.ExprInspection \
+// RUN:   -analyzer-config eagerly-assume=false -verify=x87 %s
+// RUN: %clang_analyze_cc1 -triple aarch64-unknown-linux-gnu \
+// RUN:   -analyzer-checker=core,debug.ExprInspection \
+// RUN:   -analyzer-config eagerly-assume=false -verify=quad %s
+// RUN: %clang_analyze_cc1 -triple x86_64-pc-windows-msvc \
+// RUN:   -analyzer-checker=core,debug.ExprInspection \
+// RUN:   -analyzer-config eagerly-assume=false -verify=ldbl64 %s
+// RUN: %clang_analyze_cc1 -triple powerpc64le-unknown-linux-gnu \
+// RUN:   -analyzer-checker=core,debug.ExprInspection \
+// RUN:   -analyzer-config eagerly-assume=false -verify=ibm128 %s
----------------
steakhal wrote:

I think none of the `eagerly-assume=false` are needed. And I'm curious of the verify prefixes. Where did those names come from?

In repeated cases frequently valuable to use the `DEFINE` directive to hoist out the repeated parts of the `RUN` lines.

https://github.com/llvm/llvm-project/pull/214098