| 1 | #!/usr/bin/env bash
|
| 2 | #
|
| 3 | # What do Unix tools do with "bad" filenames?
|
| 4 | #
|
| 5 | # - Those with invalid unicode
|
| 6 | # - Those with terminal escape sequences
|
| 7 | #
|
| 8 | # Usage:
|
| 9 | # data_lang/quoting-survey.sh <function name>
|
| 10 |
|
| 11 | set -o nounset
|
| 12 | set -o pipefail
|
| 13 | set -o errexit
|
| 14 |
|
| 15 | # We already know:
|
| 16 | # - bash ${#len} operator is very broken
|
| 17 |
|
| 18 | # in bash it could be %Q. Or maybe it's %Q everywhere?
|
| 19 | # in mycpp, we can translate %r calling repr() to qsn::encode()
|
| 20 |
|
| 21 |
|
| 22 | # Summary:
|
| 23 | #
|
| 24 | # These tools do UTF-8 error decoding, but they use a funny shell-like format:
|
| 25 | #
|
| 26 | # - GNU coreutils: ls, cp, stat
|
| 27 | # - GNU tar
|
| 28 | # - zsh in error message, and in printf %q
|
| 29 | # - bash and mksh in printf %q only, but not in error message
|
| 30 | # - GNU findutils: find, but NOT xargs
|
| 31 | #
|
| 32 | # Surprise: not grep
|
| 33 | #
|
| 34 | # TODO: automate this a bit?
|
| 35 | # - You can validate their stdout and stderr?
|
| 36 | # - Look for the literal escape sequence.
|
| 37 |
|
| 38 |
|
| 39 | # TODO: What about the one that changes the title?
|
| 40 |
|
| 41 | BOLD=$'\x1b[1m'
|
| 42 | RESET=$'\x1b[0;0m'
|
| 43 |
|
| 44 | # A mix of valid and invalid utf-8
|
| 45 | char_then_byte=$'\xce\xce\xbc'
|
| 46 | byte_then_char=$'\xce\xbc\xce'
|
| 47 |
|
| 48 | readonly DIR=_tmp/qsn-demo
|
| 49 |
|
| 50 | banner() {
|
| 51 | echo
|
| 52 | echo =====
|
| 53 | echo "$@"
|
| 54 | echo
|
| 55 | }
|
| 56 |
|
| 57 | setup-bad-files() {
|
| 58 |
|
| 59 | # - Make a file with an invalid code point, and utf-8 char
|
| 60 | # - Make a file with a terminal escape sequence
|
| 61 |
|
| 62 | mkdir -p $DIR
|
| 63 | pushd $DIR
|
| 64 | touch -- $BOLD $char_then_byte $byte_then_char
|
| 65 | }
|
| 66 |
|
| 67 | test-programs() {
|
| 68 | echo "$BOLD Hello $RESET World"
|
| 69 |
|
| 70 | # does approximate decoding
|
| 71 | printf '%q\n' "$char_then_byte"
|
| 72 | printf '%q\n' "$byte_then_char"
|
| 73 |
|
| 74 | setup-bad-files
|
| 75 | # ls doesn't print these by default, that' sgood
|
| 76 |
|
| 77 | # Hm this also does approximate decoding
|
| 78 | banner 'ls'
|
| 79 | ls
|
| 80 | echo
|
| 81 | ls --escape
|
| 82 | echo
|
| 83 | # Test out error message
|
| 84 | # It's basically correct, but ugly. There are too many segments, and
|
| 85 | # there's an unnecessary leading ''.
|
| 86 | # J8 is shorter and more consistent.
|
| 87 |
|
| 88 | ls -- "$RESET" || true
|
| 89 |
|
| 90 | # same
|
| 91 | banner 'cp'
|
| 92 | cp -- "$RESET" /tmp || true
|
| 93 |
|
| 94 | # weird output but it ultimately understands it
|
| 95 | banner 'stat'
|
| 96 | stat *
|
| 97 |
|
| 98 | # Hm also understands utf-8
|
| 99 | banner 'find'
|
| 100 | find
|
| 101 | # This prints it raw
|
| 102 | #find -print0
|
| 103 |
|
| 104 | # xargs --verbose messes up! Makes it bold. It also understands less
|
| 105 | # unicode.
|
| 106 | if false; then
|
| 107 | banner 'xargs'
|
| 108 | echo * | xargs --verbose -n 1 -- true
|
| 109 | fi
|
| 110 |
|
| 111 | # prints bytes, no unicode
|
| 112 | banner 'strace'
|
| 113 | strace -- true "$BOLD" "$char_then_byte" "$byte_then_char"
|
| 114 |
|
| 115 | # it does understand mu
|
| 116 | banner 'ps'
|
| 117 | bash -c "true zzmagic $BOLD $char_then_byte $byte_then_char; sleep 2" &
|
| 118 | ps aux | grep zzmagic
|
| 119 | }
|
| 120 |
|
| 121 | test-errors() {
|
| 122 | # also prints it
|
| 123 | setup-bad-files
|
| 124 |
|
| 125 | # GOOD
|
| 126 | banner 'tar'
|
| 127 | tar -f $BOLD || true
|
| 128 | tar --create "$BOLD" "$byte_then_char" "$char_then_byte" > out.tar
|
| 129 | tar --list < out.tar
|
| 130 |
|
| 131 | banner 'rm'
|
| 132 | # works
|
| 133 | rm -f -v -- "$BOLD" "$byte_then_char" "$char_then_byte"
|
| 134 |
|
| 135 | banner 'grep'
|
| 136 | # BUG
|
| 137 | #grep z "$BOLD"
|
| 138 | grep z "$byte_then_char" || true
|
| 139 | grep z "$char_then_byte" || true
|
| 140 |
|
| 141 | # python doesn't print it somehow?
|
| 142 | banner 'python'
|
| 143 | # BUG: Python prints terminal sequences
|
| 144 | #python "$BOLD" || true
|
| 145 | python "$byte_then_char" || true
|
| 146 | python "$char_then_byte" || true
|
| 147 |
|
| 148 | # BUG: Lua prints terminal sequences
|
| 149 | # So coreutils does it right!
|
| 150 | banner 'lua'
|
| 151 | #lua "$BOLD" || true
|
| 152 | lua "$byte_then_char" || true
|
| 153 | lua "$char_then_byte" || true
|
| 154 |
|
| 155 | # BUG: prints it
|
| 156 | banner 'awk'
|
| 157 | #awk -F "$BOLD" || true
|
| 158 | awk -F "$byte_then_char" || true
|
| 159 | awk -F "$char_then_byte" || true
|
| 160 |
|
| 161 | # BUG
|
| 162 | banner 'ruby'
|
| 163 | #ruby "$BOLD" || true
|
| 164 | ruby "$byte_then_char" || true
|
| 165 | ruby "$char_then_byte" || true
|
| 166 |
|
| 167 | # BUG
|
| 168 | banner 'perl'
|
| 169 | #perl "$BOLD" || true
|
| 170 | perl "$byte_then_char" || true
|
| 171 | perl "$char_then_byte" || true
|
| 172 |
|
| 173 | # BUG
|
| 174 | # But it's a little smarter about mu cases
|
| 175 | banner 'nodejs'
|
| 176 | #nodejs "$BOLD" || true
|
| 177 | nodejs "$byte_then_char" || true
|
| 178 | nodejs "$char_then_byte" || true
|
| 179 |
|
| 180 | # shells:
|
| 181 |
|
| 182 | # BUG
|
| 183 | banner 'bash'
|
| 184 | #bash "$BOLD" || true
|
| 185 | bash "$byte_then_char" || true
|
| 186 | bash "$char_then_byte" || true
|
| 187 |
|
| 188 | banner 'dash'
|
| 189 | #dash "$BOLD" || true
|
| 190 |
|
| 191 | # zsh actually escapes it!
|
| 192 | banner 'zsh'
|
| 193 | zsh "$BOLD" || true
|
| 194 | zsh "$byte_then_char" || true
|
| 195 | zsh "$char_then_byte" || true
|
| 196 |
|
| 197 | # BUG
|
| 198 | banner 'mksh'
|
| 199 | #mksh "$BOLD" || true
|
| 200 | }
|
| 201 |
|
| 202 | test-busybox() {
|
| 203 | setup-bad-files
|
| 204 |
|
| 205 | # displays ?? -- doesn't understand unicode
|
| 206 | banner 'busybox ls'
|
| 207 | busybox ls
|
| 208 |
|
| 209 | # BUG: prints it literally
|
| 210 | banner 'busybox find'
|
| 211 | busybox find
|
| 212 |
|
| 213 | #reset
|
| 214 | }
|
| 215 |
|
| 216 | "$@"
|