[FFmpeg-devel] [PATCH v5 1/4] libavcodec/dnxuc_parser: DNxUncompressed essence parser

[martin schitter]

On 11.09.24 21:47, Marton Balint wrote:
> The order of the patches is wrong. Every point in a series must be 
> compilable, and a patch in a series must only depend on earlier patches 
> in the series.
> 
> So as a first patch you should add codec ID and codec desc, then you can 
> add MXF demuxer support, then the parser and the decoder. And when 
> adding the parser/decoder, also make the needed changes in the Makefile, 
> not just as a separate patch.

You're right! I already recognized this defect yesterday when I saw the 
patchwork results. :(

I'm really unhappy about all this troubles related to git jugging and 
commit history modification.

On one hand I understand the demand for better reviewable small chunks 
of code -- and really appreciate all enormous valuable feedback here on 
the list! --, but I also see the inevitable drawbacks of this kind of 
workflow.

I'm rather sure, that nobody had recognized, that beside tiny fixes and 
rearrangements of this last patch set one file simple got lost.

I therefore personally prefer to use git version control not so much as 
a utility to create fancy code rearrangements and cosmetic cleanup, but 
as tool to control immutable history and traceable actual code changes, 
which it was made for.

In case of horrible sloppy developers like me, this may even uncover and 
blame questionable professional qualities by just unforgiving document 
all this frequently corrected typos and ugly glitches. But at the end 
it's still a rather useful instrument to correct and overcome exactly 
this kind of flaws.

I'm therefor unsure, if I'll really follow your advice and waste even 
more time on this error-prone code juggling game or just squash all the 
changes again to a reliable working bundle satisfying the final 
dependencies?

These more structured little code components, which will always compile 
if applied one after the other, may indeed look useful and seducing, but 
they would have to work even if applied in reverse order if they really 
represent individual pickable entities.

Well -- I'll see what I can do to make everybody happy...

Martin