This change makes it much easier to have a certain database administrator peer or cert authentication, that allows connecting as any user. Without this change you would need to add a line to pg_ident.conf for every user that is in the database.
In some small sense this is a breaking change if anyone is using "all" as a user currently and has pg_ident.conf rules for it. This seems unlikely, since "all" was already handled specially in pg_hb.conf. Also it can easily be worked around by quoting the all token in pg_ident.conf. As long as this is called out in the release notes it seems okay to me. However, if others disagree there would be the option of changing the token to "pg_all". Since any pg_ prefixed users are reserved by postgres there can be no user. For now I used "all" though to stay consistent with pg_hba.conf.
+1 from me. I recently was setting up a Vagrant VM for testing and wanted to allow the OS user which runs the application to connect to the database as whatever user it wants and was surprised to find I had to list all the potential target DB users in the pg_ident.conf (in production it uses password authentication and each server gets just the passwords it needs stored in ~/.pgpass). I like the idea that both config files would be consistent, although the use of keywords such as "replication" in the DB column has always made me a bit uncomfortable.
Related question: is there a reason why pg_ident.conf can't/shouldn't be replaced by a system table? As far as I can tell, it's just a 3-column table, essentially, with all columns in the primary key. This latest proposal changes that a little; strictly, it should probably introduce a second table with just two columns identifying which OS users can connect as any user, but existing system table style seems to suggest that we would just use a special value in the DB user column for "all".
