build: Only allow ASCII identifiers #19094

While emoji and other symbols in C++ identifers (as accepted by newer compilers) are fun, they might create confusion during code review, for example because some symbols look very similar. Forbid such extended identifiers for now.

This is done by providing -fno-extended-identifiers. Thanks to sipa for suggesting this compiler flag.

ACK, though as long as we have at least a GCC before 10.0 in our CI, such identifiers won't work regardless.

Note that this flag also does not exist for clang (which seems to have no way of disallowing such identifiers, or non-ascii input in general).

Concept ACK. The same is enforced for the python functional tests (emojis in symbol names are not allowed by python, but strings may contain arbitrary utf-8 characters).

Edit: It looks like python3 does allow some non-ascii unicode: https://docs.python.org/3/reference/lexical_analysis.html#identifiers

Sorry, I meant ℭ𝔬𝔫𝔠𝔢𝔭𝔱 𝔄ℭ𝔎

ACK, though as long as we have at least a GCC before 10.0 in our CI, such identifiers won't work regardless.

You're right with regard to CI. Then it's mostly useful for people developing locally, I suppose. I tend to prefer problems being flagged while compiling locally to having to PR and wait for travis to fail.

ACK 399d84da3708719b063953107bab0f5f6493ad

sanity check that -fno-extended-identifiers was not present in build configure on master and is present in this branch

checking whether C++ compiler accepts -fno-extended-identifiers... yes

  CC            = /usr/bin/ccache gcc
  CFLAGS        = -g -O2
  CPPFLAGS      =   -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=2  -DHAVE_BUILD_INFO -D__STDC_FORMAT_MACROS
  CXX           = /usr/bin/ccache g++ -std=c++11
  CXXFLAGS      =   -fstack-reuse=none -Wstack-protector -fstack-protector-all
                    -Wall -Wextra -Wformat -Wvla -Wswitch -Wredundant-decls
                    -Wunused-variable -Wdate-time -Wsign-compare
                    -Wno-unused-parameter -Wno-implicit-fallthrough  
                    -g -O2 -fno-extended-identifiers
  LDFLAGS       = -pthread  -Wl,-z,relro -Wl,-z,now -pie  
  ARFLAGS       = cr

edit: gcc version 8.3.0 on debian

The gcc 10 release notes kindly include an example to test against https://gcc.gnu.org/gcc-10/changes.html

gcc 10:

scheduler.cpp:74:18: error: stray ‘\317’ in program
   74 | static const int π = 3;
      |                  ^
scheduler.cpp:75:11: error: stray ‘\303’ in program
   75 | int get_naïve_pi() {
      |           ^
scheduler.cpp:76:10: error: stray ‘\317’ in program
   76 |   return π;
      |          ^
scheduler.cpp:74:21: error: expected unqualified-id before ‘=’ token
   74 | static const int π = 3;
      |                    ^
scheduler.cpp:75:13: error: expected initializer before ‘ve_pi’
   75 | int get_naïve_pi() {
      |            ^~~~~

clang 10: compiles emoji and utf-8 without warning

ACK 399d84da3708719b063953107bab0f5f6493addb -- patch looks correct

ACK 399d84da3708719b063953107bab0f5f6493addb.

ACK 399d84da3708719b063953107bab0f5f6493addb - seems like a good sanity check to enable.