This is needed so that LLD wouldn't bind a symbol to a framework where it's hidden (would result in "missing symbol" runtime crash).
Perf impact seemed minimal:
(linking chromium)
x base.txt
+ with_hide.txt
SYSTEM CPU time: 
    N           Min           Max        Median           Avg        Stddev
x   5          0.88           0.9          0.89         0.892  0.0083666003
+   5          0.87           0.9          0.88         0.882   0.010954451
No difference proven at 95.0% confidence
USER CPU time: 
    N           Min           Max        Median           Avg        Stddev
x   5          3.51          3.54          3.51         3.518   0.013038405
+   5          3.47           3.5           3.5          3.49   0.014142136
Difference at 95.0% confidence
	-0.028 +/- 0.019837
	-0.795907% +/- 0.56387%
	(Student's t, pooled s = 0.0136015)
WALL time: 
    N           Min           Max        Median           Avg        Stddev
x   5          3.91          3.94          3.94          3.93   0.014142136
+   5          3.86           3.9           3.9         3.888   0.017888544
Difference at 95.0% confidence
	-0.042 +/- 0.0235167
	-1.0687% +/- 0.598389%
	(Student's t, pooled s = 0.0161245)PR/52708
nit: looks like this could go above the bool assignments