Snakemake cascading checkpoints not working - nested

Question

I have following situation
Checkpoint A creates files (either 1 or 2) in folder A_Out
Checkpoint B takes all the files from A_Out as an input and generates files (1 or 2) in B_Out
I have conditional rule C which can take input from A or B and produce final output
I am using following code,
checkpoint A:
output: directory("A_Out")
# Rest of the logic
def collect_a(wildcards):
path = checkpoints.A.get(**wildcards).output[0]
return expand("{path}/{file}.txt",
path=path,
file=glob_wildcards(os.path.join(path, "{name}.txt")).name)
checkpoint B:
input: collect_a
output: directory("B_Out")
# Rest of the logic
def collect_b(wildcards):
path = checkpoints.B.get(**wildcards).output[0]
return expand("{path}/{file}.txt",
path=path,
file=glob_wildcards(os.path.join(path, "{name}.txt")).name)
def conditional_input(wildcards):
if condition_A:
return collect_a(wildcards)
else:
return collect_b(wildcards)
rule c:
input: conditional_input
# Rest of the logic
In above case, when condition_a is False, it evaluates only checkpoint B and do not evaluate checkpoint A. How can I solve this problem? Or is there any other elegant way?

Answer 1

These cascading checkpoint calls are pretty buggy and have several issues associated with them. If you flesh out your example more you can submit an issue to github, assuming you have the most recent version of snakemake.
As a workaround, you can try to add in a temporary signal file in each directory to indicate the rule has completed. You then need to do your glob wildcards in the rule code.
Not great, but could work. Because you have a temp input as a signal, the rules will rerun every time even if the other files are already present; snakemake doesn't know they are outputs. Another option would be to examine your inputs beforehand to decide if one or two files would be generated and tie that into your input function logic, bypassing checkpoints.