default punkcheck hash algorithm to md5 if sha1 unavailable or unaccelerated

src/bootsupport/modules/punk/ansi-0.1.1.tm

@@ -6439,6 +6439,8 @@ tcl::namespace::eval punk::ansi::class {
             set o_gx0states     [list]
             set o_splitindex    [list]
 
+            #sha1 takes *much* longer to compute than md5 if tcllibc not available - otherwise it is generally faster
+            #we should fall back to md5 if no acceleration available. check for command sha1::sha1c ?
             set o_cksum_command [list sha1::sha1 -hex]
 
 

src/bootsupport/modules/punk/args-0.2.tm

@@ -876,7 +876,7 @@ tcl::namespace::eval punk::args {
 
         set cache_key $args
         #ideally we would use a fast hash algorithm to produce a short key with low collision probability.
-        #something like md5 would be ok (this is non cryptographic) - but md5 uses open and isn't usable by default in a safe interp. (sha1 often faster on modern cpus)
+        #something like md5 would be ok (this is non cryptographic) - but md5 uses open and isn't usable by default in a safe interp. (sha1 often faster on modern cpus but terribly slow without an accelerator)
         #review - check if there is a built-into-tcl way to do this quickly
         #for now we will just key using the whole string
         #performance seems ok - memory usage probably not ideal

src/bootsupport/modules/punk/mix/base-0.1.tm

@@ -430,7 +430,16 @@ namespace eval punk::mix::base {
 
         #not just used by cksum_path. used by caller (e.g fill_relativecksums_from_base_and_relativepathdict via cksum_filter_opts) to determine what opt names passed through
         variable cksum_default_opts
-        set cksum_default_opts [dict create -cksum_content 1 -cksum_meta auto -cksum_acls 0 -cksum_usetar auto -cksum_algorithm sha1]
+        set default_hash sha1 ;#but fall back to md5 if either sha1 is unavailable or unaccelerated (pure tcl sha1 is way slower than pure tcl md5 - can take minutes on even moderate sized source files)
+        if {![catch {package require sha1}]} {
+            set impls [::sha1::Implementations]
+            if {[llength $impls] == 1} {
+                set default_hash md5
+            }
+        } else {
+            set default_hash md5
+        }
+        set cksum_default_opts [dict create -cksum_content 1 -cksum_meta auto -cksum_acls 0 -cksum_usetar auto -cksum_algorithm $default_hash]
         proc cksum_default_opts {} {
             variable cksum_default_opts
             return $cksum_default_opts
@@ -438,9 +447,9 @@ namespace eval punk::mix::base {
 
         #crc::cksum is extremely slow in tcllib as at 2023 e.g 20x slower (no c implementation?)
         # - try builtin zlib crc instead?
-        #sha1 is performant - and this is not being used in a cryptographic or adversarial context - so performance and practical unlikelihood of accidental collisions should be the main consideration.
+        #sha1 is performant (when accelerator present) - and this is not being used in a cryptographic or adversarial context - so performance and practical unlikelihood of accidental collisions should be the main consideration.
         #adler32 is fastest for some larger files of a few MB but slower on small files (possibly due to Tcl-based file load?)
-        #sha1 as at 2023 seems a reasonable default
+        #sha1 as at 2023 seems a reasonable default - (but only if accelerator present)
         proc cksum_algorithms {} {
             variable sha3_implementation
             #sha2 is an alias for sha256

src/bootsupport/modules/textblock-0.1.3.tm

@@ -95,6 +95,7 @@ tcl::namespace::eval textblock {
     #NOTE sha1, although computationally more intensive, tends to be faster than md5 on modern cpus
     #(more likely to be optimised for modern cpu features?)
     #(This speed improvement may not apply for short strings)
+    #This is probably only true if tcllibc is available - pure-tcl sha1 is excruciatingly slow... 
 
     variable use_hash ;#framecache
     set use_hash none ;#slightly faster but uglier layout for viewing frame_cache display

src/modules/punk/ansi-999999.0a1.0.tm

@@ -6439,6 +6439,8 @@ tcl::namespace::eval punk::ansi::class {
             set o_gx0states     [list]
             set o_splitindex    [list]
 
+            #sha1 takes *much* longer to compute than md5 if tcllibc not available - otherwise it is generally faster
+            #we should fall back to md5 if no acceleration available. check for command sha1::sha1c ?
             set o_cksum_command [list sha1::sha1 -hex]
 
 

src/modules/punk/args-999999.0a1.0.tm

@@ -876,7 +876,7 @@ tcl::namespace::eval punk::args {
 
         set cache_key $args
         #ideally we would use a fast hash algorithm to produce a short key with low collision probability.
-        #something like md5 would be ok (this is non cryptographic) - but md5 uses open and isn't usable by default in a safe interp. (sha1 often faster on modern cpus)
+        #something like md5 would be ok (this is non cryptographic) - but md5 uses open and isn't usable by default in a safe interp. (sha1 often faster on modern cpus but terribly slow without an accelerator)
         #review - check if there is a built-into-tcl way to do this quickly
         #for now we will just key using the whole string
         #performance seems ok - memory usage probably not ideal

src/modules/punk/mix/base-0.1.tm

@@ -430,7 +430,16 @@ namespace eval punk::mix::base {
 
         #not just used by cksum_path. used by caller (e.g fill_relativecksums_from_base_and_relativepathdict via cksum_filter_opts) to determine what opt names passed through
         variable cksum_default_opts
-        set cksum_default_opts [dict create -cksum_content 1 -cksum_meta auto -cksum_acls 0 -cksum_usetar auto -cksum_algorithm sha1]
+        set default_hash sha1 ;#but fall back to md5 if either sha1 is unavailable or unaccelerated (pure tcl sha1 is way slower than pure tcl md5 - can take minutes on even moderate sized source files)
+        if {![catch {package require sha1}]} {
+            set impls [::sha1::Implementations]
+            if {[llength $impls] == 1} {
+                set default_hash md5
+            }
+        } else {
+            set default_hash md5
+        }
+        set cksum_default_opts [dict create -cksum_content 1 -cksum_meta auto -cksum_acls 0 -cksum_usetar auto -cksum_algorithm $default_hash]
         proc cksum_default_opts {} {
             variable cksum_default_opts
             return $cksum_default_opts
@@ -438,9 +447,9 @@ namespace eval punk::mix::base {
 
         #crc::cksum is extremely slow in tcllib as at 2023 e.g 20x slower (no c implementation?)
         # - try builtin zlib crc instead?
-        #sha1 is performant - and this is not being used in a cryptographic or adversarial context - so performance and practical unlikelihood of accidental collisions should be the main consideration.
+        #sha1 is performant (when accelerator present) - and this is not being used in a cryptographic or adversarial context - so performance and practical unlikelihood of accidental collisions should be the main consideration.
         #adler32 is fastest for some larger files of a few MB but slower on small files (possibly due to Tcl-based file load?)
-        #sha1 as at 2023 seems a reasonable default
+        #sha1 as at 2023 seems a reasonable default - (but only if accelerator present)
         proc cksum_algorithms {} {
             variable sha3_implementation
             #sha2 is an alias for sha256

src/modules/textblock-999999.0a1.0.tm

@@ -95,6 +95,7 @@ tcl::namespace::eval textblock {
     #NOTE sha1, although computationally more intensive, tends to be faster than md5 on modern cpus
     #(more likely to be optimised for modern cpu features?)
     #(This speed improvement may not apply for short strings)
+    #This is probably only true if tcllibc is available - pure-tcl sha1 is excruciatingly slow... 
 
     variable use_hash ;#framecache
     set use_hash none ;#slightly faster but uglier layout for viewing frame_cache display