In article , "nicolas" writes: > "Larry Kilgallen" a �crit dans le > message news: jGvPbYAYRtuu@eisner.encompasserve.org... >> > Beside that, the timestamp method cannot always be trusted. >> > We had problems with gnatmake which doesn't always recompile necessary >> > files. >> > The problem occurs when several files in different directories have same >> > name, same timestamp, and are selected with ADA_INCLUDE_PATH >> > timestamp and name of the file is checked, they are the sames, gnatmake >> > doesn't figure out that this is a different file and doesn't recompile > it. >> >> Well that sounds like just an implementation bug rather than a >> conceptual flaw. >> > > We recompiled gnatmake with debug messages to see what went wrong actually. > But I'm not sure you can call that an implementation bug. > The flaw is to consider that 2 files with same name and same timestamp have > the same content, > or moreover that the content of a single file has not changed, if the > timestamp has not changed > A lot of batch, or cvs import, commit or extractions, can generate a lot of > files with same name and same timestamp. But unless they are the same bits on the disk, they are a different file. On VMS you would use the 6 byte file ID and the 64 byte device ID to determine uniqueness. I am sure the same construct must exist on other operating systems. Granted, someone writing a tool may neglect to consider the situation you describe, but that is still just an implementation bug.