|
24 | 24 | # |
25 | 25 | # WHY A SANDBOX. A machine that has been developing these packages has every one |
26 | 26 | # of them installed, and would answer for its own state rather than for the |
27 | | -# index. `xlings subos <name> --sandbox --cmd` gives a fresh environment, and a |
28 | | -# fresh `/tmp` with it --- so nothing may be staged outside. |
| 27 | +# index. `xlings subos <name> --sandbox --cmd` gives an environment of its own, |
| 28 | +# and a `/tmp` of its own with it --- so nothing may be staged from outside. |
| 29 | +# |
| 30 | +# ⚠️ SEPARATE IS NOT FRESH, AND THIS COMMENT SAID FRESH. Measured 2026-08-27: |
| 31 | +# two invocations of the same environment, and the second found the directory |
| 32 | +# the first had made; the host's `/tmp` had neither. So the `/tmp` is not the |
| 33 | +# host's --- which is what matters for staging --- and it is not new each time, |
| 34 | +# which is what matters for a check that must not read its own last answer. |
| 35 | +# That is why the project directory below is REMOVED before it is written. |
29 | 36 | # |
30 | 37 | # It asks two questions, because they fail independently: |
31 | 38 | # |
@@ -116,6 +123,10 @@ got="$(mcpp --version | awk '{print $2}')" |
116 | 123 | echo " mcpp is $got" |
117 | 124 |
|
118 | 125 | say "a project that names only published versions" |
| 126 | +# Removed rather than merely created: see the note about `/tmp' at the head of |
| 127 | +# this file. A directory left by the previous run would be built again, and a |
| 128 | +# build that succeeded against the previous release would look exactly like one |
| 129 | +# that succeeded against this one. |
119 | 130 | rm -rf /tmp/closure && mkdir -p /tmp/closure/src && cd /tmp/closure |
120 | 131 | cat > mcpp.toml <<TOML |
121 | 132 | [package] |
|
0 commit comments