Laurent, as Eric explained, this commit (as well as the homogeneisation of spacing) are fixes to pre-existing issues in the script, in order to better match other bash sources in the codebase. It would be possible to pick 1/10 and 2/10 as a separate patchset.
El mar., 10 mar. 2020 a las 12:47, Eric Blake (<ebl...@redhat.com>) escribió: > On 3/10/20 3:28 AM, Laurent Vivier wrote: > > Le 09/03/2020 à 20:19, Unai Martinez-Corral a écrit : > >> All the tests are prefixed with 'x', in order to avoid risky comparisons > >> (i.e. a user deliberately trying to provoke a syntax error). > > > > With the quotes I don't see how we can provoke a syntax error. > > Could you provide an example? > > Historically, in some shells: > > foo=\( > bar=\) > if [ "$foo" = "$bar" ]; then echo hello world; fi > > could output 'hello world' (by parsing a parenthesized one-argument > test, and the string '=' is non-empty), but: > > if [ "x$foo" = "x$bar" ]; then echo goodbye; fi > > did not (since no operator begins with 'x', you have guaranteed the > syntax that [ will parse). Similarly, if foo=! or foo=-a, you could get > syntax errors (if [ tried to treat the expansion of $foo as an operator > and got thrown off by the remaining arguments not matching an expected > pattern). > > These days, POSIX says that with three arguments when the 2nd is a > binary operator, there is no ambiguity (the binary operator takes > precedence over the ( and ) around the non-empty string test), and > modern bash obeys the POSIX rule without needing the x prefix. But it > is still better to prefix with x for copy-paste portability to older > shells that do not match current POSIX rules. > > -- > Eric Blake, Principal Software Engineer > Red Hat, Inc. +1-919-301-3226 > Virtualization: qemu.org | libvirt.org > >
