[gentoo-pms] REPLACING_VERSIONS

[gentoo-pms] REPLACING_VERSIONS

[Ulrich Mueller]

PMS currently says: "A whitespace-separated list of versions of this
package (including revision, if specified) that are being replaced
(uninstalled or overwritten) as a result of this install."

Shouldn't the spec guarantee that it's the empty string (i.e., not
a string containing arbitrary whitespace) if the list is empty?
Otherwise simple binary tests like [[ -n ${REPLACING_VERSIONS} ]]
cannot be done.

For REPLACED_BY_VERSION, it is explicitly specified: "The single
version ..., or an empty string otherwise."

Ulrich

Re: [gentoo-pms] REPLACING_VERSIONS

[Ciaran McCreesh]

[-- Attachment #1: Type: text/plain, Size: 931 bytes --]

On Mon, 31 Jan 2011 19:23:51 +0100
Ulrich Mueller <ulm@gentoo.org> wrote:
> PMS currently says: "A whitespace-separated list of versions of this
> package (including revision, if specified) that are being replaced
> (uninstalled or overwritten) as a result of this install."
> 
> Shouldn't the spec guarantee that it's the empty string (i.e., not
> a string containing arbitrary whitespace) if the list is empty?
> Otherwise simple binary tests like [[ -n ${REPLACING_VERSIONS} ]]
> cannot be done.
> 
> For REPLACED_BY_VERSION, it is explicitly specified: "The single
> version ..., or an empty string otherwise."

We had an issue with this early on with Paludis: certain eclasses were
expecting exact spacing requirements inside A and AA based upon a
quirk in how Portage created them, and it was causing weird bugs. I'd
be inclined to say that people should write more tolerant code...

-- 
Ciaran McCreesh

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: [gentoo-pms] REPLACING_VERSIONS

[Ulrich Mueller]

>>>>> On Mon, 31 Jan 2011, Ciaran McCreesh wrote:

>> Shouldn't the spec guarantee that it's the empty string (i.e., not
>> a string containing arbitrary whitespace) if the list is empty?
>> Otherwise simple binary tests like [[ -n ${REPLACING_VERSIONS} ]]
>> cannot be done.

> We had an issue with this early on with Paludis: certain eclasses
> were expecting exact spacing requirements inside A and AA based upon
> a quirk in how Portage created them, and it was causing weird bugs.
> I'd be inclined to say that people should write more tolerant
> code...

No, nothing like exact spacing requirements in the general case. Only
simplify developers' lifes in what I guess will be 90% of the usage.

And I think that one could already read "space separated" in that way:
If there are no elements, there is nothing to separate.

Ulrich

Re: [gentoo-pms] REPLACING_VERSIONS

[Ulrich Mueller]

Anyway, another dev has also complained that the cheat sheet's wording
for these variables would not be clear enough.

A new attempt is included below, please review. (If you suggest a new
wording, I'd really like it not to exceed 5 lines in the PDF output.)

--- a/eapi-cheatsheet.tex
+++ b/eapi-cheatsheet.tex
@@ -205,9 +205,9 @@ nonetheless.
     \item[\code{PROPERTIES}] Is mandatory for all package managers now
     to support interactive installs.
     \item[\code{REPLACING\_VERSIONS}, \code{REPLACED\_BY\_VERSION}]
-    For corner cases in upgrade paths those variables can be defined
-    in \code{pkg\_*} functions to define a package version that is
-    replaced by or is the replacement for the current one.
+    These variables, valid in \code{pkg\_*}, contain a list of all
+    versions (\code{PVR}) of this package that we are replacing, and
+    the version that is replacing the current one, respectively.
     See \featureref{replace-version-vars}.
     \item[\code{MERGE\_TYPE}] This variable contains one of three
     possible values to allow checks if it is normal merge with