Performance regression with Improve printing of several arguments (#55754)

[Zentrik]

This pr made the alloc.strings benchmark in BaseBenchmarks 2000% slower in min wall time, use 42.25% more memory and increase allocations by 63.53%.

Also, it worsened compile performance with inference.abstract interpretation.Base.init_stdio(::Ptr{Cvoid}) being 742% slower and inference.allinference.Base.init_stdio(::Ptr{Cvoid}) 15% slower. There is also a slowdown recently in the println(::QuoteNode) benchmark which is probably due to this. (I'm mentioning this as the compile time impact was investigated in the PR).

Until the next nanosoldier report, data available at https://tealquaternion.camdvr.org/compare.html?start=58b239c5b8eaffec3b9b99e6d7c37e8ae6129d6d&end=6e5e87b2cafda840b90347c2e74202d2608d7c29&stat=min-wall-time.

Also, this made the sys.so 11MiB bigger on the buildkite x86_64-linux-gnu build.
Before:

section                      size       addr
.hash                         832        624
.gnu.hash                      60       1456
.dynsym                      2616       1520
.dynstr                      1876       4136
.gnu.version                  218       6012
.gnu.version_r                176       6232
.rela.dyn                 1146744       6408
.rela.plt                    2064    1153152
.init                          36    1159168
.plt                         1392    1159216
.plt.got                       16    1160608
.text                    30754273    1160624
.fini                          14   31914900
.rodata                     92380   31916032
.eh_frame_hdr              375628   32008412
.eh_frame                 1875288   32384040
.ctors                         16   34266496
.dtors                         16   34266512
.dynamic                      480   34266528
.got                          128   34267008
.got.plt                      712   34267136
.data                           8   34267848
.bss                           16   34267856
.lbss                       85872   34267872
.lrodata                   251552   34357840
.ldata                  122156256   34613488
.comment                       17          0
.debug_aranges                128          0
.debug_info              18645878          0
.debug_abbrev               32104          0
.debug_line               9379182          0
.debug_str                 329379          0
.debug_ranges            18087504          0
.debug_gnu_pubnames       1773625          0
.debug_gnu_pubtypes        180702          0
Total                   205177188

After:

/cache/build/builder-amdci5-6/julialang/julia-master/usr/lib/julia/sys.so  :
section                      size       addr
.hash                         832        624
.gnu.hash                      60       1456
.dynsym                      2616       1520
.dynstr                      1876       4136
.gnu.version                  218       6012
.gnu.version_r                176       6232
.rela.dyn                 1179240       6408
.rela.plt                    2064    1185648
.init                          36    1187840
.plt                         1392    1187888
.plt.got                       16    1189280
.text                    30449361    1189296
.fini                          14   31638660
.rodata                     92240   31641600
.eh_frame_hdr              386460   31733840
.eh_frame                 1922456   32120304
.ctors                         16   34049408
.dtors                         16   34049424
.dynamic                      480   34049440
.got                          128   34049920
.got.plt                      712   34050048
.data                           8   34050760
.bss                           16   34050768
.lbss                       84352   34050784
.lrodata                   250816   34139232
.ldata                  134215360   34394144
.comment                       17          0
.debug_aranges                128          0
.debug_info              18946308          0
.debug_abbrev               31998          0
.debug_line               9289599          0
.debug_str                 339061          0
.debug_ranges            17350624          0
.debug_gnu_pubnames       1864018          0
.debug_gnu_pubtypes        189054          0
Total                   216601768

[Zentrik]

Oddly this also seemed to regress the performance of other inference benchmarks like rand(::Float64). I can't replicate locally but doesn't seem to be noise as it persisted for multiple runs.

Significant changes in benchmarks below

Summary

	Range	Mean	Count
Regressions	7.34%, 1019.85%	448.16%	7
Improvements	0.00%, 0.00%	0.00%	0
All	0.00%, 1019.85%	448.16%	7

Benchmarks

Benchmark	% Change	Significance Factor
inference.abstract interpretation.broadcasting	1019.85%	200.75x
inference.abstract interpretation.rand(Float64)	751.83%	89.52x
inference.abstract interpretation.Base.init_stdio(::Ptr{Cvoid})	750.08%	22.32x
inference.abstract interpretation.REPL.REPLCompletions.completions	395.36%	7.78x
inference.abstract interpretation.println(::QuoteNode)	200.49%	16.09x
inference.allinference.rand(Float64)	12.17%	1.70x
inference.allinference.broadcasting	7.34%	1.11x

Source: https://tealquaternion.camdvr.org/compare.html?start=220742d6194acba995eda822c82fdf647d6896ee&end=44bef0df7a115334c10abac88aeba333b12cce2d&stat=min-wall-time&nonRelevant=true.

[vtjnash]

It is probably a cumulative effect of the memory usage from the primary breakage after running those tests in order