Unlimited (100x) undo buffer with compression #201
Conversation
|
Sorry, I have parts in the PR that were not meant to be in there. Not sure how this happens. I will tidy this up now... |
src/main.c
Outdated
| cairo_set_operator(context->paint_ctx, CAIRO_OPERATOR_ATOP); | ||
| else /* GROMIT_PEN */ | ||
| cairo_set_operator(context->paint_ctx, CAIRO_OPERATOR_OVER); | ||
| else if (type == GROMIT_RECOLOR) |
| void snap_undo_state (GromitData *data) | ||
| { | ||
| if(data->debug) | ||
| g_printerr ("DEBUG: Snapping undo buffer %d.\n", data->undo_head); | ||
|
|
||
| copy_surface(data->undobuffer[data->undo_head], data->backbuffer); | ||
| undo_compress(data, data->backbuffer); |
There was a problem hiding this comment.
Why not have compression and saving in one function?
There was a problem hiding this comment.
...because it is a swap operation which requires three steps, and these steps are combined differently in snap_.... and in undo/redo...
In snap_undo_state we have
undo_compress(data, data->backbuffer); // data -> temp[LZ4]
undo_temp_buffer_to_slot(data, data->undo_head); // temp[LZ] -> undo_slot[LZ4]
In undo/redo we have the 3-step swapping:
undo_compress(data, data->backbuffer); // data -> temp[LZ4]
undo_decompress(data, data->undo_head, data->backbuffer); // undo_slot[i,LZ4] -> data
undo_temp_buffer_to_slot(data, data->undo_head); // temp[LZ4] -> undo_slot[i,LZ4]
The other aspect is that temp_buffer (step 1) is generously grown when it is too small (it is doubled), whereas the undo_buffers are only every grown to the size needed. The doubling is not critical because there is only one instance of temp_buffer, whereas there are 100 instances of the undo_buffer
src/main.h
Outdated
| @@ -193,4 +201,5 @@ void paint_context_free (GromitPaintContext *context); | |||
|
|
|||
| void indicate_active(GromitData *data, gboolean YESNO); | |||
|
|
|||
This (small) PR implements LZ4 compression of the undo buffers.
LZ4 compression/decompression is very fast, and with the drawings I tested, achieves a compression down to 1.5-2.5% of the original buffer. Hence, this allows for a very deep undo buffer without significant memory penalty.
I have fixed it at 100 which for practical purposes is unlimited.
The entire logic of the undo system remains unaltered.
The only difference is that
swap_surfaceincludes a compression/decompression step.The undo buffers are only allocated once they are needed, and they are successively grown when this is necessary.