> Huh. I'd say the examples are perfectly good and on-point. While dealing with autoexec.bat and random BSODs wasn't fun, it's entirely orthogonal to the fact that a DOS-era POS still offers orders of magnitude better UX than current-era browser POSes, or than most web apps for that matter.

I know of an ERP system that somehow manages to take about 15 seconds to search an inventory of ~100k items. If you export all those items to CSV, with all their attributes (most of which are not searched), the resulting file is about 15 MB.

It is boggling how they managed to implement search this slowly (in a C++ application using MS SQL as the backend). 3.5 GHz computers, performing plain text string search at about 1 MB/s.

It is even more surprising that users feel this is not an completely unreasonable speed.

(They managed to pull this stunt off by completely not using the SQL database in the intended way, i.e. all tables are essentially (id, blob) tuples, where the blob is a zlib compressed piece of custom TLV encoded data. All data access goes through a bunch of stored procedures, which return data in accordance to a sort of "data extraction string". Search works by re-implementing inverted indices in tables of (word, offset, blob), where blob contains a zlib compressed list of matching IDs; again processed by stored procedures. The client then is wisely implemented using MS SQL's flavour of LIMIT queries which effectively cause a classic quadratic slowdown because the database engine literally has no way to fetch result rows n...m except by constructing the entire result set up to m.

Unsurprisingly the developers of this abomination claim to be competent. They also invented a funny data exchange format involving fixed field lengths and ASCII separators - some time in the 2010s.)