On 5/16/19 12:36 AM, Martin Sebor wrote: > On 5/15/19 8:50 AM, Martin Sebor wrote: >> On 5/15/19 5:40 AM, Martin Liška wrote: >>> On 5/14/19 11:31 PM, Martin Sebor wrote: >>>> The attached patch implements the -Wformat-diag warning to help find >>>> quoting, spelling, and other formatting issues in diagnostics issued >>>> by GCC. >>> >>> Just a general comment about this part. Wouldn't it make sense to use regex >>> for some of the string patterns that you address in the patch? >> >> I'm not sure. Are you wondering about the variants of type >> names like "signed int" and "unsigned int" and the contractions? >> >> The plain format checker scans and advances one word at a time >> to be efficient. It does linearly loop over the black-listed >> words and operators and that could be made more efficient by >> using a binary search. Let me do that. I would expect >> introducing regexp to the checking to slow it down but >> I suppose I could try it and see if you think it's worth it. > > I tried both approaches for the operators and keywords:
Hi. Thanks for considering. It was just an idea I did. > > 1) using bsearch > 2) using regex > > I finished (1) but did only a proof of concept of (2). Neither > made a noticeable difference in my (admittedly very simple) > benchmarks but both have drawbacks: > > (1) means that the entries in the tables of known operators and > keywords have to be sorted. For operators, the sort order is > different between ASCII and EBCDIC, and would mean duplicating > the operator table and making sure it stays sorted in each > encoding after a change. That seems like a pain in the butt > for little gain. > > In my prototype of (2), the regular expression quickly started > to get complicated and hard to read as I tried to handle some > of the same special cases as I'm handling now. It might be > possible to redo the whole thing using regex but it would mean > starting from scratch. I'm not convinced it's worth the effort > (in fact, I would worry about the regular expressions becoming > too complex to easily extend). > > So I'm inclined to leave things as they are. I did make a few > simplifications to the code in the process of this experiment > and beefed up the test a little bit so I attach an updated > revision. Works for me. I'm going to test your patchset on ppc64le and I'll prepare patch for the target. Martin > > Martin
