On Sat, Nov 28, 2020 at 09:14:35PM +0100, Pavel Stehule wrote: > Any short or long option can be read from this file in simple format - one > option per line. Arguments inside double quotes can be multi lined. Row > comments started by # and can be used everywhere.
here is updated patch
Does this support even funkier table names ?
This tests a large number and fraction of characters in dbname/username, so all of pg_dump has to continue supporting that: ./src/bin/pg_dump/t/010_dump_connstr.pl
I tested and it seems to work with -t "fooå" But it didn't work with -t "foo\nbar" (literal newline). Fix attached. If you send another patch, please consider including a test case for quoted names in long and short options.
I implemented some basic backslash escaping. I will write more tests, when there will be good agreement on the main concept.
> +static char *optsfilename = NULL;
> + * It assign the values of options to related DumpOption fields or to > + * some global values. It is called from twice. First, for processing > + * the command line argumens. Second, for processing an options from > + * options file.
This didn't support multiple config files, nor config files which include config files, as Dean and I mentioned. I think the argument parsers should themselves call the config file parser, as need be, so the last option specification should override previous ones.
For example pg_dump --config-file=./pg_dump.conf --blobs should have blobs even if the config file says --no-blobs. (Command-line arguments normally take precedence over config files, certainly if the argument is specified "later"). I think it'd be ok if it's recursive. I made a quick hack to do that.
I did it. I used a different design than you. Making "dopt" be a global variable looks too invasive. Almost all functions there expect "dopt" as an argument. But I think it is not necessary.
I implemented two iterations of argument's processing. 1. for options file (more options-file options are allowed, and nesting is allowed too), 2. all other arguments from the command line. Any options file is processed only once - second processing is ignored. So there is no problem with cycles.
The name of the new option - "config-file" or "options-file" ? I prefer "options-file". "config-file" is valid too, but "options-file" is more specific, more descriptive (it is self descriptive).
I merged your patch with a fix of typos.
Regards
Pavel
I doubt this will satisfy Stephen. Personally, I would use this if it were a plain and simple text config file (which for our purposes I would pass on stdin), and I would almost certainly not use it if it were json. But it'd be swell if there were a standard config file format, that handled postgresql.conf and maybe pg_hba.conf.
