Test failures on ppc64le

[QuLogic]

Running on ppc64le, the 1.12.6 build fails (note this is a scratch build that will be deleted, so I will copy as much as possible here) on 4 tests.

Error in test.data.table() : 
  4 errors out of 8132 in 00:01:56 elapsed (00:04:13 cpu) on Thu Nov  7 09:58:44 2019.
 [endian==little, sizeof(long double)==16, sizeof(pointer)==8, TZ=NA, locale='C', l10n_info()='MBCS=FALSE; UTF-8=FALSE; Latin-1=FALSE', getDTthreads()='omp_get_num_procs()==12; R_DATATABLE_NUM_PROCS_PERCENT==unset (default 50); R_DATATABLE_NUM_THREADS==unset; omp_get_thread_limit()==2147483647; omp_get_max_threads()==12; OMP_THREAD_LIMIT==unset; OMP_NUM_THREADS==unset; RestoreAfterFork==true; data.table is using 6 threads. See ?setDTthreads.'].
 Search tests/tests.Rraw.bz2 for test numbers: 1018, 1729.08, 1729.11, 1729.13.

I cat'd the main.Rout.fail to get the full error text:

Running test id 1018      Test 1018 ran without errors but failed check that x equals y:
> x = identical(fread("1.46761e-313\n"), data.table(V1 = 1e-313)) 
First 1 of 1 (type 'logical'): 
[1] FALSE
> y = TRUE 
First 1 of 1 (type 'logical'): 
[1] TRUE
1 element mismatch

Running test id 1729.08      Test 1729.08 ran without errors but failed check that x equals y:
> x = x 
First 6 of 505 (type 'character'): 
[1] "-6e-05"            "-123456789.123457" "-1000"            
[4] "-983"              "-966"              "-949"             
> y = y 
First 6 of 505 (type 'character'): 
[1] "-6e-05"            "-123456789.123457" "-1000"            
[4] "-983"              "-966"              "-949"             
2 string mismatches
   row            fwrite              write.csv
1: 502  5.123456789e-307  5.12345678900000e-307
2: 503 -5.123456789e-307 -5.12345678900000e-307

Running test id 1729.11      Test 1729.11 did not produce correct output:
Expected: <<V1\n5.123456789e+300\n-5.123456789e+300\n1e-305\n1e+305\n1.2e-305\n1.2e+305\n1.23e-305\n1.23e+305>>
Observed: <<V1\n5.123456789e+300\n-5.123456789e+300\n1.00000000000000e-305\n1e+305\n1.20000000000000e-305\n1.2e+305\n1.23000000000000e-305\n1.23e+305>>

Running test id 1729.13      Test 1729.13 ran without errors but failed check that x equals y:
> x = capture.output(fwrite(DT)) 
First 5 of 5 (type 'character'): 
[1] "V1"                    "2.22044604925031e-16"  "1.11022302462516e-16" 
[4] "2.2250738585072e-308"  "1.79769313486232e+308"
> y = capture.output(write.csv(DT, row.names = FALSE, quote = FALSE)) 
First 5 of 5 (type 'character'): 
[1] "V1"                    "2.22044604925031e-16"  "1.11022302462516e-16" 
[4] "2.22507385850720e-308" "1.79769313486232e+308"
1 string mismatch

The first one looks like #3492 but the last 3 seem to be an annoying, but benign, difference in whether trailing zeros are printed.

Output of sessionInfo()

Sorry, not really available on the builder, but you can take a look at the root.log if necessary.

[QuLogic]

For 1729, I worked around it by adding a gsub("\\.?0+e", "e", ). Don't know if that's the way you want to go.

[QuLogic]

For 1018, it seems that the difference is in fread:

> 1.46761e-313
[1] 1.467610e-313
> as.double("1.46761e-313")
[1] 1.467610e-313

> library(data.table)
data.table 1.12.6 using 1 threads (see ?getDTthreads).  Latest news: r-datatable.com

> fread("1.46761e-313\n")
              V1
1: 1.467608e-313

> data.table(V1=1.46761e-313)
              V1
1: 1.467610e-313

though I'm not sure why the printout from the tests is showing data.table to be different.

Note that the exponent factor here should be 1.0E-318, which does not show up as a truncated to zero warning in the lookup table. Truncation to zero starts at -324 and overflow starts at +309.