In T898#14811, @nrwahl2 wrote:

In T898#14801, @kgaillot wrote:

Since the risk of corner cases is so high, and our time for 3.0.0 is limited, I think the best approach would be to log a warning if a CIB contains ACLs with xpaths and the upgrade adds or renames any element or attribute. (We could possibly check whether an xpath exists that refers to an attribute, to reduce the scope a bit.) Something like: "WARNING: CIB syntax changes may invalidate ACLs that use 'xpath'. It is strongly recommended to run 'cibadmin --upgrade' then go through the updated CIB carefully to ensure ACLs still match the desired intent."

I'd also be fine with warning at the end if any ACLs use xpath, period. It's broader scope than necessary, but it avoids the need to clutter the stylesheet and the logs with individual warnings for each relevant transformation. (We could some of the log clutter by having one warning per step in the pipeline, but that may be a gnarly template to catch every case in a single template.)

Since the risk of corner cases is so high, and our time for 3.0.0 is limited, I think the best approach would be to log a warning if a CIB contains ACLs with xpaths and the upgrade adds or renames any element or attribute. (We could possibly check whether an xpath exists that refers to an attribute, to reduce the scope a bit.) Something like: "WARNING: CIB syntax changes may invalidate ACLs that use 'xpath'. It is strongly recommended to run 'cibadmin --upgrade' then go through the updated CIB carefully to ensure ACLs still match the desired intent."

In T896#14685, @nrwahl2 wrote:

XSL has some mechanism for this, and we already have an appropriate logger in C.

Logging a warning via XSL (using xsl:message) looks straightforward. I presume libxslt will send it to stderr... It's not obvious how we'd tell the C caller to log a message, or what it should log. Maybe we could do some before-and-after checking of the CIB XML, but that doesn't seem worth the computational cost or added code complexity.

Fixed by rPc1f3ae1d9f (accidentally put wrong task number in commit message)