> So in conclusion we have 3 possible paths to go with this > > 1) Just don't allow any extensions referenced by other extensions to be > relocatable. > It will show a message something like > "SET SCHEMA not allowed because other extensions depend on it" > Given that if you don't specify relocatable in you .control file, the assume is > relocatable = false , this isn't too far off from standard protocol. > > 2) Use objsubid=1 to denote that another extension explicitly references the > schema of another extension so setting schema of other extension is not okay. > So instead of introducing another dependency, we'd update the > DEPENDENCY_NORMAL one between the two schemas with objsubid=1 > instead of 0. > > This has 2 approaches: > > a) Update the existing DEPENDENCY_NORMAL between the two extensions > setting the objsubid=1 > > or > b) Create a new DEPEDENCY_NORMAL between the two extensions with > objsubid=1 > > I'm not sure if either has implications in backup / restore . I suspect b would > be safer since I suspect objsubid might be checked and this dependency only > needs checking during SET SCHEMA time. > > 3) Create a whole new DEPENDENCY type, perhaps calling it something like > DEPENDENCY_EXTENSION_SCHEMA > > 4) Just don't allow @extschema:<reqextension>@ syntax to be used unless > the <reqextension> is marked as relocatable=false. This one I don't like > because it doesn't solve my fundamental issue of > > postgis_tiger_geocoder relying on fuzzystrmatch, which is marked as > relocatable. > > The main issue I was trying to solve is my extension references fuzzystrmatch > functions in a function used for functional indexes, and this fails restore of > table indexes because I can't schema qualify the fuzzystrmatch extension in > the backing function. > > > If no one has any opinion, I'll go with option 1 which is the one that strk had > actually proposed before and seems least programmatically invasive, but > perhaps more annoying user facing. > > My preferred would be #2 > > Thanks, > Regina
Attached is my revision 3 patch, which follows the proposed #1. Don't allow schema relocation of an extension if another extension requires it.
0003-Allow-use-of-extschema-reqextname-to-reference.patch
Description: Binary data
